🏛️ Issues and Advocacy Training™
Courses › Cybersecurity for Regulated and Large Organizations: Foundations™

Lesson 9 of 12 · about 55 minutes · Updated November 14, 2026

People: Training, Phishing, Social Engineering and Insider Risk

You'll be able to

  • Explain what large field studies from 2022 to 2025 found about security awareness training and phishing simulations, and which measures matter more than click rates
  • Design a help desk identity verification procedure and payment verification rules that resist voice, video and deepfake impersonation
  • Describe the parts of an insider risk program using CISA's four steps and a cross-discipline team, and the general legal limits on employee monitoring
  • Recognize the warning signs of North Korean remote IT worker schemes and apply the FBI's 2025 hiring checklist
  • Apply background check rules, separation of duties and joiner, mover and leaver controls at the scale of a large organization
  • Write an acceptable use policy and build a reporting culture that rewards fast reporting instead of punishing honest mistakes

Start here: the short version

  • Most breaches still pass through a person. Verizon's 2025 Data Breach Investigations Report found a human element in 60 percent of the breaches it studied, and the FBI's Internet Crime Complaint Center logged 191,561 phishing and spoofing complaints in 2025.12
  • Annual slide-deck training on its own does little. Two of the largest field studies (Lain, Kostiainen and Capkun, 2022; Ho and colleagues, 2025) found that common training formats barely changed click rates. What does move: making it fast and easy to report a suspicious message, short role-based training for the jobs criminals actually target, and technical controls such as phishing-resistant sign-in that make a click less costly.34
  • Criminals now aim at the people who can reset passwords and move money: help desks, payroll, accounts payable and executive assistants. The 2020 Twitter hack, the 2023 MGM and Clorox attacks and the 2024 Arup deepfake video call each turned on a person being talked into something.56
  • An insider risk program covers careless and malicious insiders alike. CISA's model is four steps (define, detect and identify, assess, manage), run by a team that includes HR, security, IT, compliance and management, with clear limits on what is monitored and why.7
  • North Korean "remote IT workers" using stolen or borrowed U.S. identities have been hired by hundreds of U.S. companies. One Arizona facilitator alone helped place workers at 309 U.S. companies and was sentenced to 102 months in July 2025. Hiring and identity checks are now a security control.8
  • Monitoring employees is lawful in most U.S. workplaces when it is disclosed and tied to a business purpose, but the limits vary by state, by union status and by whether the employer is a government. Write the rules down, tell people, and keep the program proportionate.
  • A culture that rewards fast reporting and does not punish honest mistakes catches more attacks than a culture of fear. This lesson ends with a sample Acceptable Use Policy.

Rules, standards and reports described: FTC Safeguards Rule, 16 CFR 314.4(e) (as amended through the 2023 notification amendment, effective May 13, 2024); HIPAA Security Rule, 45 CFR 164.308(a)(5) (current rule; the January 2025 proposed update was still a proposal when checked); NYDFS 23 NYCRR 500.14 (as amended Nov. 1, 2023); NIST SP 800-50 Rev. 1 (September 2024), SP 800-53 Rev. 5, SP 800-171 Rev. 3 (May 2024) and SP 800-46 Rev. 2 (July 2016); PCI DSS v4.0.1 (described, not quoted); 32 CFR 117.7 (insider threat programs for cleared contractors); CISA Insider Threat Mitigation Guide (November 2020; updated edition September 2026); FBI public service announcements I-012325-PSA, I-051525-PSA and I-121925-PSA; FinCEN Alert FIN-2024-Alert004 (Nov. 13, 2024); Verizon 2025 Data Breach Investigations Report; FBI Internet Crime Report 2025 (released April 2026); New York Civil Rights Law 52-c; the Fair Credit Reporting Act as explained by the FTC. All read or checked Nov. 13 to 14, 2026. Every number is labeled with its source year. Laws, regulations and guidance change; check the regulator's current text before relying on any point.

Before you change anything. Plan changes to production systems, test them first, keep a way back, and follow your change management process. For clinical, plant-floor or other safety systems, involve the vendor, your safety lead and qualified engineers (see modules 109 and 111).

On the afternoon of July 14, 2020, a handful of Twitter employees working from home took phone calls from someone who said he was with the company's IT help desk. Twitter had moved to remote work a few months before, and VPN trouble was common, so a call about a VPN problem sounded normal. The caller steered each employee to a login page that looked exactly like Twitter's real VPN page, on a look-alike domain. As the employees typed their usernames, passwords and one-time codes, the attackers replayed them in real time. By the afternoon of July 15 the intruders, led by a 17-year-old in Florida, had reached internal account tools and taken over 130 accounts, including those of a former president, a presidential candidate and major companies. They posted a bitcoin scam that brought in about $118,000. New York's Department of Financial Services, which investigated because cryptocurrency firms regulated in New York were impersonated, wrote that the whole attack happened in about 24 hours.5

No exploit was needed. No firewall failed. A person on a phone call was the way in. That pattern repeats through most of the large breaches of the last decade, and it is why every serious security program for a hospital, a bank, a law firm, a manufacturer, a city or a public company treats people as part of the security system, not as a problem to be lectured once a year.

This lesson is about that part of the system. It covers what the research actually shows about training and phishing simulations, how social engineering has moved to phones, help desks and synthetic voices and video, how to build an insider risk program that is fair and lawful, how North Korean remote worker schemes have turned hiring into a security question, and how remote work, acceptable use rules, background checks, separation of duties and a reporting culture fit together. It is written for executives and board members, compliance officers, HR leaders, CISOs and IT staff, help desk managers, finance and payroll leads, and the reporters, advocates and public officials who ask how organizations protect the people whose data they hold.

If you want the small and medium business version first, People, Policies and Training in module 73 covers the eight-policy set, hiring and the Fair Credit Reporting Act, offboarding and a first 90 days at small-business scale, and Scams Aimed at Businesses in module 72 walks through the scams themselves. This lesson does not repeat those. It goes further, at the scale of organizations with hundreds or thousands of people, regulators watching, and help desks and HR departments big enough to be targets in their own right.

Inside this module, the sign-in controls that make people harder to fool are in lesson 5, Identity and Access; the logging that lets an insider risk team see anything at all is in lesson 10, Detection and Incident Response; the board's oversight of the whole program is in lesson 4; and outside contractors and staffing firms, who are insiders too, are in lesson 8, Vendors and Supply Chain.

Why people are the target

Attackers go where the work is easiest. As organizations have closed open ports, patched faster and put endpoint detection on laptops, the cheapest way in has often become a person: someone who will type a password into a fake page, approve a sign-in prompt they did not start, reset an account for a stranger, or wire money because the boss seemed to ask for it.

What the breach data shows

Verizon's Data Breach Investigations Report (DBIR) is the most widely cited annual count of breaches, built from cases supplied by dozens of investigators, insurers and government agencies. The 2025 edition, covering incidents mostly from late 2023 through late 2024, found "human involvement" in 60 percent of breaches, compared with 61 percent the year before, a difference Verizon says is within its margin of error. The human element in Verizon's sense covers social engineering, errors such as sending data to the wrong person, misuse of privileges, and the use of stolen credentials. Verizon notes that the share has drifted down a little since 2022, mainly because criminals are exploiting more software vulnerabilities in edge devices, not because people got harder to fool.1

The same report found that when the actor was an insider, unintentional mistakes outnumbered deliberate misuse about two to one. Most of the insiders were ordinary end users, and their most common mistake was misdelivery: sending an email, file or paper document to the wrong recipient, which made up 72 percent of end-user error actions.1 That is a training and process problem, not a hacking problem, and it is one of the most common reasons organizations end up sending breach notices under state laws such as Idaho's.

What the FBI complaint data shows

The FBI's Internet Crime Complaint Center (IC3) takes complaints from victims at ic3.gov. Its 2025 report, released in April 2026, counted 1,008,597 complaints and $20.877 billion in reported losses, a 26 percent increase in losses over 2024.2 Several of the categories describe crimes that begin with a person being deceived:

  • Phishing and spoofing: 191,561 complaints in 2025, the largest category by count, with reported losses of $215,843,126.
  • Business email compromise: $3,046,598,558 in reported losses, the second-largest loss category, built almost entirely on fooling someone in finance or operations. (Module 110 covers it in depth in Business Email Compromise and Wire Fraud.)
  • Tech support scams: 47,794 complaints and $2,134,675,818 in losses.
  • Government impersonation: 32,424 complaints and $797,943,193 in losses.
  • Employment scams: 24,688 complaints and $362,934,762 in losses.

For the first time the 2025 report also gave a separate count of complaints that mentioned artificial intelligence: 22,364 complaints with adjusted losses of $893,346,472. Within that, businesses reported more than $30 million in losses to business email compromise involving AI, and victims reported almost $13 million in losses to AI-involved employment scams. The report describes job interviews in which the lip movements of the person on camera did not match the audio, and says the goal of those schemes "generally appears to be gaining access to private computer networks."9

Two cautions apply to IC3 numbers. They count only what victims chose to report, so they are a floor. And a company that loses money to an insider or a fake employee often does not file with IC3 at all. The numbers show direction and scale, not the whole problem.

What the regulators require

Nearly every security rule that reaches a regulated organization requires training, and several now name social engineering or insider threats specifically. The table below lists the main ones. It describes them in plain terms; the exact wording is in each rule.

Rule or standard (version)Who it coversWhat it says about people
FTC Safeguards Rule, 16 CFR 314.4(e) (amended rule fully in force since June 9, 2023; notice amendment effective May 13, 2024)Non-bank financial institutions: mortgage brokers, auto dealers, tax preparers, payday lenders, many fintechs and othersProvide personnel with security awareness training "updated as necessary to reflect risks identified by the risk assessment," and use qualified information security personnel who get security updates and training.10
HIPAA Security Rule, 45 CFR 164.308(a)(5)Covered entities and business associatesImplement a security awareness and training program for all members of the workforce, including management, with addressable specifications for security reminders, protection from malicious software, log-in monitoring and password management.11
NYDFS Cybersecurity Regulation, 23 NYCRR 500.14 (as amended Nov. 1, 2023)Banks, insurers and other entities licensed by the New York Department of Financial ServicesCybersecurity awareness training for all personnel at least annually, including social engineering, updated to reflect the entity's risk assessment; plus risk-based monitoring of authorized users' activity.12
NIST SP 800-171 Rev. 3 (May 2024), requirements 03.02.01 and 03.02.02Contractors holding Controlled Unclassified Information, through DFARS and CMMCLiteracy training at onboarding and periodically, including recognizing and reporting signs of insider threat, social engineering and social mining, plus role-based training.13
NIST SP 800-53 Rev. 5, controls AT-2, AT-2(2), AT-2(3), AT-3, PM-12Federal systems and anyone who adopts the catalog, including many states and contractorsAwareness training, with enhancements for insider threat and for social engineering and mining; role-based training; and an insider threat program with a cross-discipline team.14
PCI DSS v4.0.1 (June 2024), Requirement 12.6Anyone who stores, processes or transmits payment card dataA formal security awareness program at hire and at least once every 12 months; requirements that became mandatory on March 31, 2025 add awareness of phishing and related social engineering attacks and acceptable use of end-user technologies. (Described here; the standard itself is copyrighted.)15
32 CFR 117.7 (the National Industrial Security Program Operating Manual rule)Contractors cleared to handle classified informationAn insider threat program led by a named Insider Threat Program Senior Official, with training, under Executive Order 13587.16

None of these rules says how good the training must be, and none says "do phishing simulations." That leaves the design to the organization, which is where the research matters.

Screenshot of a French-language email that is an attempt to phish an email account, from 2011.
A French email account phishing attempt from 2011. The tricks have changed less than people think: messages like this typically claim to come from a system administrator, give a reason to act now, and link to a form that collects the password. Most organizations still receive messages like this every day, and the question this lesson asks is what actually reduces the damage when one gets through. Photo: Phishing killer, 2011-01-19. Wikimedia Commons, Public domain.

Security awareness training that works

Security awareness training is a large industry, and most regulated organizations buy some version of it: an annual video course with a quiz, plus monthly simulated phishing emails that send anyone who clicks to a short lesson. For years the industry measured success by falling click rates in its own simulations. Independent researchers have now tested those claims in real organizations at large scale, and the results should change how boards and compliance officers spend training money.

What the research says

Lain, Kostiainen and Capkun (ETH Zurich), 2022. The researchers worked with a large company for 15 months (July 2019 to October 2020), sending eight simulated phishing emails to each of more than 14,000 employees, 117,864 simulated phishes in all, and randomly assigning people to different conditions. Overall, 32.10 percent of participants clicked at least one simulated phish and 25.43 percent did something dangerous at least once, such as entering credentials. The most striking finding concerned "embedded" training, the common practice of showing a short lesson to anyone who clicks: people in the group that got it clicked more, not less (3,593 clicks with training versus 3,087 without), and more of them became repeat clickers. Warnings on suspicious emails helped, but detailed warnings were no better than short ones. The study also found that employees, given an easy report button, became a fast and useful detection system: the company received 18,476 reports, and about 10 percent of reports on a simulated campaign arrived within five minutes of it landing.3

Ho, Mirian, Luo and colleagues (UC San Diego and UC San Diego Health), 2025. This team ran a randomized experiment for eight months across more than 19,500 employees of a large health system, with ten simulated phishing campaigns. They found no meaningful relationship between when someone had last completed the mandated annual training and whether they fell for a phish. Embedded training reduced the chance of clicking by about 2 percent, a difference the authors called extremely low. Most people barely looked at the training page: in more than 75 percent of training sessions people spent less than a minute on it, and 37 to 51 percent of sessions showed no engagement at all. About 10 percent of employees clicked in the first month, but more than half had clicked at least once by month eight. The lure mattered more than the training: only 1.82 percent clicked an email about updating an Outlook password, while 30.8 percent clicked one about a change to the health system's vacation policy. The authors concluded that training programs in their current forms "are unlikely to offer significant practical value in reducing phishing risks" and pointed to two-factor authentication and password managers that fill in passwords only on the correct website as better investments.417

Verizon, 2025. The DBIR examined phishing simulation data from security awareness vendors. Click rates did not fall to zero even after years of training, with roughly 1.5 percent of employees in the median organization still clicking; Verizon suggested there may be a ceiling on what education can do. But reporting moved a great deal: employees who had training within the previous 30 days reported simulated phishing at about 21 percent, against a base rate of about 5 percent, a fourfold increase. The effect of recent training on clicking was far smaller.18

NIST, 2020. Researchers at the National Institute of Standards and Technology (Michelle Steves, Kristen Greene and Mary Theofanos) published the "Phish Scale," a method for rating how hard a simulated phish is to spot, based on how many warning cues it has and how well it fits the recipient's work. Their point was simple and important: a click rate means nothing unless you know how hard the test was. A program that switches from obvious lures to well-targeted ones will see click rates rise even as people get better.19

What large studies found about phishing training Bars drawn to scale, 0 to 60 percent. Each group is a different study; do not compare across groups. 0% 20% 40% 60% Lain et al., 2022 (14,000+ staff, 15 months) Clicked at least once (8 tests) 32.10% Dangerous action at least once 25.43% Ho et al., 2025 (19,500+ staff, 8 months) Clicked in first month about 10% Clicked at least once by month 8 over 50% Embedded training cut clicking by only about 2 percent. Verizon DBIR 2025 (vendor simulation data) Report rate, no recent training about 5% Report rate, trained in last 30 days about 21% The pattern: training moved reporting far more than it moved clicking.
Three large studies of phishing training, side by side. Original graphic for this module; data from Lain, Kostiainen and Capkun (IEEE Symposium on Security and Privacy, 2022), Ho and colleagues (IEEE Symposium on Security and Privacy, 2025) and the Verizon 2025 Data Breach Investigations Report.

What to take from the research

None of these studies says to stop training. They say to stop expecting training alone to keep people from clicking, and to measure the right things. Five conclusions hold up across the evidence:

  1. Someone will always click. In every large study a meaningful share of people fell for at least one realistic lure. A security program that fails when one person clicks is a program designed to fail. The real defenses sit behind the click: phishing-resistant multifactor authentication (MFA, a second proof of identity beyond a password; the phishing-resistant kinds are FIDO2 security keys, passkeys and smart cards such as the federal PIV card, which will not hand a code to a fake site), email filtering, endpoint detection and response (EDR) software on every computer, and least-privilege access so one account cannot reach everything. CISA's 2022 fact sheet on implementing phishing-resistant MFA explains how these methods work and how to roll them out in stages. Those controls are covered in lesson 5 and lesson 6.20
  2. Reporting is the measure that matters. A phishing campaign aimed at 2,000 employees is caught when the first few report it and the security team pulls the message from every inbox. Lain and colleagues showed that crowd reporting is fast and accurate enough to be a real detection layer, and Verizon found that recent training quadrupled reporting.318
  3. Embedded "gotcha" training is weak. Two independent studies found it did little or made things slightly worse. If you use simulations, use them mainly to practice reporting and to find where real lures would succeed, not to punish.
  4. Context beats volume. A vacation-policy lure fooled 30.8 percent of health system employees while a password-reset lure fooled 1.82 percent.4 The lures that work are the ones that look like the job. Training should therefore be built around each role's real risks: invoices for accounts payable, résumés with attachments for recruiters, password-reset calls for the help desk, board packets for executive assistants.
  5. Know how hard your tests are. Use the NIST Phish Scale or a similar method so that leaders do not mistake easier tests for better people.19

Building a learning program the NIST way

NIST's guidance on training was rewritten in September 2024 as SP 800-50 Rev. 1, Building a Cybersecurity and Privacy Learning Program. It replaced documents from 2003 and 1998 and, for the first time, treats privacy and security together. It describes a life cycle with four phases (plan and strategy; analysis and design; development and implementation; assessment and improvement) and builds content with the familiar ADDIE model: analysis (identify learning needs from each work role's tasks, the audience, their current knowledge and the gaps), design, development, implementation and evaluation. It explicitly aims at changing behavior as part of risk management, not at completing modules, and it asks programs to set metrics and evaluate them.21

For a large organization that turns into a short list of practical choices:

  • Short and frequent beats long and annual. A ten-minute module each quarter tied to current scams, plus brief reminders when a real campaign is hitting the sector, keeps the habit alive. The annual course can stay to satisfy rules that require it, but do not mistake it for the program.
  • Role-based tracks. Everyone gets the basics. Higher-risk roles get more: finance and payroll (payment verification, vendor bank changes), the help desk and IT administrators (identity verification for resets, refusing remote-access requests), HR and recruiters (fake applicants, malicious attachments, identity checks), executives and their assistants (impersonation, deepfakes, travel), developers (secrets in code, package risks), clinicians and plant operators (the systems they run and who to call), and anyone who handles regulated data (misdelivery, minimum necessary).
  • Teach reporting first. Every course should end with one action: how to report, in one click or one call, and what happens next.
  • Train leaders separately. Board members and senior executives are named targets of impersonation and the people whose behavior sets the culture. A 30-minute briefing a year, ideally built around a tabletop exercise (see lesson 10), does more than putting them in the general course.
  • Include contractors and temporary staff. Rules such as HIPAA and NYDFS cover all personnel or workforce members, not only employees, and attackers do not check payroll status.

Metrics boards should see

Completion rates are what most dashboards show, because they are easy to count. They mostly prove that a rule was followed. More useful measures, which a board or audit committee can ask for (see lesson 4 on board metrics):

MeasureWhat it tells youWatch out for
Report rate on simulated and real phishingWhether staff act as sensorsReports of harmless mail; make reporting easy anyway
Time from first delivery to first report, and to removal from all inboxesHow fast the organization contains a real campaignNeeds a reporting button wired to the security team's tools
Real incidents first spotted by an employeeWhether the people program catches real attacksRequires incident records to note who spotted it
Help desk resets that passed the verification procedure, and spot-check resultsWhether the highest-risk human process is being followedNeeds call recording or ticket evidence
Payment changes verified by callback before releaseWhether finance follows the anti-fraud ruleSample real transactions, not self-reports
Click rate, with test difficulty notedTrend in susceptibility, if difficulty is held steadyMeaningless without the difficulty rating
Accounts with phishing-resistant MFA, by risk tierHow much a click can costExceptions for executives and service accounts

Phishing simulations, done fairly

Simulations are not useless. They let a security team see which lures would work, give people practice with the report button, and give a regulator or auditor evidence that the program is active. But they are a deception aimed at your own staff, and how they are run decides whether they build trust or destroy it.

  • Do not use lures that exploit employees' livelihoods or fears. Fake bonuses, pay raises, layoff notices or health scares produce high click rates and lasting resentment, and they teach people that the security team plays tricks. The vacation-policy result in the UC San Diego study shows workplace topics are effective lures; that is a reason to warn about them, not to use the cruelest versions on your own people.
  • Tell staff the program exists. A simulation program announced in policy is a training exercise. One run in secret is a trap.
  • Coordinate with HR, the privacy office and any union. A simulation that collects data about individual employees is a form of monitoring; where workers are represented, it may be a subject of bargaining.
  • Thank reporters, every time. An automatic "thanks, that was a test" message, and a human thank-you when a report catches a real attack, make reporting the normal thing to do.
  • Handle repeat clickers with support, not punishment. Look at why: an overloaded role, a job that requires opening attachments from strangers, poor tools. Often the fix is technical (a sandbox for attachments, a separate inbox for public submissions), not disciplinary.
  • Exclude real crisis periods. Do not run tests during a real incident, a merger announcement or a natural disaster, when a fake message can cause real harm.

CISA, NSA, the FBI and MS-ISAC's October 2023 joint guide on phishing puts training alongside technical defenses (email authentication, filtering, phishing-resistant MFA, blocking malicious macros) rather than in place of them, which is the right order.22

Social engineering beyond email

Social engineering means manipulating a person into doing something that helps an attacker: revealing a password, approving a sign-in, resetting an account, installing remote-access software, or sending money. Email is still the most common channel, but the most damaging attacks of recent years have used the telephone, text messages, collaboration apps and, increasingly, synthetic voices and video. Verizon's 2025 report found that phishing and pretexting (inventing a believable story to get someone to act) remain the main social engineering methods, and that "synthetically generated text" in malicious emails had doubled over two years, according to one of its data partners.23

The help desk is the front door

In a large organization, the IT service desk can reset any employee's password and re-enroll their multifactor device. That makes it the most valuable human target in the building. Three well-documented cases show the pattern.

Twitter, 2020. As described at the top of this lesson, the attackers posed as Twitter's own help desk, calling remote employees about VPN problems and sending them to a fake VPN page. New York regulators found the attack went from first calls to account takeovers in roughly 24 hours, and noted that remote work had made VPN complaints routine, which made the pretext believable.5

The "Scattered Spider" attacks, 2022 to 2025. CISA and the FBI first described this loose group of criminals (widely reported to be mostly young and English-speaking) in a joint advisory on Nov. 16, 2023, and updated it on July 29, 2025, with partners in Canada, Australia and the United Kingdom, based on FBI investigations through June 2025. The updated advisory says the actors posed as company IT or help desk staff to get employees' credentials or to direct them to run remote-access tools, and "posed as employees to convince IT and/or helpdesk staff to provide sensitive information, reset the employee's password, and transfer the employee's MFA to a device they control." It also says they "abuse trusted relationships of contracted IT help desks." Most recently they deployed DragonForce ransomware. The advisory's list of actions to take today includes enforcing phishing-resistant MFA, and it says organizations "should continue to perform diligent employee training against vishing" (voice phishing).24

MGM Resorts, 2023. Attackers associated with the same group disrupted MGM's casinos and hotels in September 2023. In a Form 8-K filed Oct. 5, 2023, MGM said it expected the attack to have a negative impact of about $100 million on its adjusted property earnings for the third quarter, plus less than $10 million in one-time expenses. Press accounts at the time reported that the intruders talked their way past the IT help desk; MGM's filing does not describe how they got in, so treat that detail as reported, not confirmed.25

Clorox, 2023, and the lawsuit that followed. In July 2025, Clorox sued Cognizant, which ran Clorox's IT service desk under a contract dating to 2013, in California state court, seeking $380 million. Clorox's complaint alleges that on Aug. 11, 2023, an attacker posing as an employee called the service desk and that agents reset the person's password and multifactor credentials without the identity verification Clorox's procedures required. Cognizant has denied responsibility and said publicly that it was hired for a narrow help desk role and "did not manage cybersecurity for Clorox." In March 2026 the judge dismissed one claim (intentional misrepresentation) but let most of the case proceed. These are allegations in a pending case, not findings; the point for every organization is that a service desk script, and whether staff follow it, can be worth hundreds of millions of dollars.2627

The lesson from all three is the same. The identity check at the help desk must be strong enough to stop a confident stranger who knows the employee's name, title, manager, employee number and the last four digits of a Social Security number, because all of that is for sale or findable. A safe procedure looks like this:

  1. Never verify identity using facts an attacker can buy. Name, date of birth, employee ID, manager's name and partial SSN are not proof.
  2. Prefer self-service with strong proof. Let employees reset their own passwords through a portal that requires a registered phishing-resistant authenticator, so most resets never reach a person.
  3. For calls the desk must handle, call back. End the call and call the employee at the number in the HR system of record, not a number the caller gives. Or verify on video against the photo on file, or in person at a badge office.
  4. Get a second confirmation for high-risk accounts. For administrators, executives, finance staff and anyone with access to regulated data, require the employee's manager (reached by a known channel) to confirm before any MFA re-enrollment.
  5. Never reset MFA and password in the same call. Changing both at once hands over the whole account. Require a waiting period or a separate verified step for MFA changes.
  6. Tell the real owner. Send an automatic alert to the employee's existing email and phone, and to the security team, whenever a password or MFA device changes.
  7. Hold the outsourced desk to the same rule in writing. Put the verification procedure in the contract, require recordings or ticket evidence, and audit samples monthly (see lesson 8 on vendor contracts).
  8. Give agents permission to say no. The caller who claims to be a senior vice president, angry and in a hurry, is the classic pretext. Make it a rule, backed by leadership, that no one is punished for following the procedure.

Tightening reset procedures changes how thousands of people get back into their accounts. Plan the change, test it with a pilot group and have a way back if the new process locks people out of critical systems.

Help desk reset: a safe path Caller asks for a password or MFA reset Can they use self-service with a registered phishing-resistant authenticator? yes Employee resets it themselves. Done. no End the call. Call back at the HR system's number, or verify on video against the photo, or in person. Verified? no Refuse, log, alert security yes High-risk account? (admin, executive, finance, regulated data) yes Manager confirms by a known channel Reset one factor per verified step, never both at once Alert the employee's old email, phone and security Record evidence in the ticket for monthly audit Not proof of identity: name, birth date, employee ID, manager's name, last four of an SSN.
A help desk reset procedure that would have stopped the pretexts described in the CISA and FBI Scattered Spider advisory. Original graphic for this module; based on CISA and FBI advisory AA23-320A (updated July 29, 2025) and the NYDFS Twitter report (Oct. 14, 2020).

Voice, video and deepfake fraud

In January 2024, an employee in the Hong Kong office of Arup, the British engineering firm, received a message that appeared to come from the company's UK-based chief financial officer about a confidential transaction. The employee then joined a video conference in which the CFO and several colleagues appeared and spoke. All of them were digital recreations. Over about a week the employee made 15 transfers totaling HK$200 million (about US$25.6 million) to five bank accounts. Hong Kong police described the case publicly in February 2024 without naming the company and said it was the first time they had seen a fake multi-person video conference used this way. Arup confirmed in May 2024 that it was the victim and said none of its internal systems had been compromised.6

That last sentence is the point. No system was hacked. The company's payment controls trusted what an employee saw and heard on a video call. U.S. agencies have since issued several warnings:

  • FinCEN Alert FIN-2024-Alert004 (Nov. 13, 2024) told banks and credit unions that criminals were using "deepfake media" created with generative AI to get around identity verification, open accounts with fake documents and photos, and commit fraud, and listed red flags such as inconsistencies in identity documents and customers who will not use multifactor checks.28
  • FBI PSA I-051525-PSA (May 15, 2025) warned of a campaign since April 2025 using text messages and AI-generated voice messages that impersonated senior U.S. officials to build rapport, move targets to other platforms and steal account access. Its advice: "do not assume it is authentic," verify by calling a number you find independently, and never share a two-factor code.29
  • FBI PSA I-121925-PSA (Dec. 19, 2025) updated that warning: since at least 2023 the impersonators had posed as senior state officials, White House and Cabinet officials and members of Congress, contacted their families and acquaintances, and asked for authentication codes, passport copies, overseas wires and introductions.30
  • The FBI's 2025 IC3 report counted 22,364 AI-related complaints with $893,346,472 in losses, including more than $30 million in AI-involved business email compromise.9

Public officials in Idaho and every other state should read the 2025 FBI warnings as aimed at them: county commissioners, agency directors, legislators and their staff are exactly the kind of senior officials and acquaintances described.

Verification rules that beat deepfakes

Detecting a good deepfake by eye or ear is unreliable and getting harder; the IC3 report says AI-enabled synthetic content is "becoming increasingly difficult to detect."9 The defense is procedural: a rule that certain actions never happen on the strength of a single channel, no matter who appears to be asking.

RequestThe rule
New or changed bank details for a vendor, client, employee or payrollCall back at a number from the vendor master file or contract, never one in the request; a second person approves.
Urgent or confidential payment requested by an executiveNo payment on any single channel, including live video. Confirm through a separate, pre-agreed channel, and follow the normal approval chain. No executive may waive it.
Password or MFA resetThe help desk procedure above.
Request to install remote-access software or run a commandOnly from a ticket opened through the official system; staff hang up and call the service desk number they already know.
Request for a sign-in code or push approvalNever share a code or approve a prompt you did not start. No legitimate IT staff will ask.
Request for employee W-2s, payroll files or patient listsTreat as high risk; verify by callback and require a second approval from the data owner.
Executive or official contacted by "a colleague" on a new number or appConfirm through a known channel before replying with anything sensitive; consider a code word for executive teams and families, as the FBI suggests.

Write these rules down, have the CEO and CFO sign them, and tell staff that the executives themselves will be tested. A rule that leaders exempt themselves from is the rule criminals exploit.

Insider risk programs

An insider is anyone who has or had authorized access. CISA defines an insider as "any person who has or had authorized access to or knowledge of an organization's resources," and insider threat as "the threat that an insider will use their authorized access, wittingly or unwittingly, to do harm."31 The word "unwittingly" matters. Most insider damage is accidental, as Verizon's two-to-one ratio of errors to deliberate misuse shows, and a program that only hunts for traitors will miss most of the risk and alienate the workforce.

Kinds of insiders

Type (CISA's terms)What it looks likeMain controls
NegligentCareless handling: emailing patient files to a personal account to work at home, sharing a password, ignoring a rule to save timeClear rules, tools that make the safe way easy, data loss prevention (DLP) that warns before sending, training
AccidentalHonest mistakes: misdelivered email, a misconfigured cloud share, a lost laptopAuto-complete limits for external addresses, delay-send for sensitive mail, encryption, default-private sharing
Intentional (malicious)Theft of data to sell or take to a competitor, sabotage by a departing or disgruntled employee, fraudLeast privilege, separation of duties, monitoring of high-risk access, fast offboarding, an assessment team
CollusiveInsiders working with outside criminals, often for bribesLimits on what any one support role can see, monitoring for bulk lookups, vendor oversight
Third partyContractors, vendors and staffing firm workers with accessContract terms, the same access rules as employees, offboarding at contract end (see lesson 8)

Type names and definitions from CISA's "Defining Insider Threats" page.31

A recent case: bribed support agents

In May 2025, Coinbase, the largest U.S. cryptocurrency exchange, disclosed in a Form 8-K that criminals had bribed a small number of customer support agents working for an overseas contractor to copy customer data from internal support tools. The data included names, addresses, phone numbers, email addresses, masked Social Security and bank account numbers, images of government ID and account balances. The criminals used it to impersonate Coinbase and trick customers into sending them funds, then demanded $20 million on May 11, 2025, to keep quiet. Coinbase refused, offered a $20 million reward for information instead, said it had fired the agents involved, and estimated its costs for remediation and voluntary customer reimbursement at $180 million to $400 million.32

The lessons are concrete. Support roles need customer data, but not all of it, and not in bulk. Tools should show the minimum needed for each call, flag agents who look up far more accounts than their tickets explain, and treat contractor staff as insiders. The case is also a reminder that insider risk and vendor risk are the same problem when the insider works for a vendor. Verizon's 2025 report noticed the same shift, with "an increase in Partner actors" in its privilege misuse cases and a rise in breaches caused by system administrators, whose access is far broader than ordinary users'.33

Building the program

CISA's Insider Threat Mitigation Guide (first published November 2020; updated edition September 2026, which adds material on AI-enabled deception, hybrid and remote work, access control and involuntary separations) is the best free starting point for any organization, public or private, and is written to scale from small agencies to large companies. It organizes the work into four steps: defining the threat, detecting and identifying it, assessing it and managing it.7 In late January 2026, CISA added an infographic, "Assembling a Multi-Disciplinary Insider Threat Management Team," aimed at critical infrastructure and local governments.34 For federal contractors with security clearances, a program is not optional: 32 CFR 117.7 requires a cleared contractor to "establish and maintain an insider threat program" under Executive Order 13587, led by an Insider Threat Program Senior Official appointed in writing.16 NIST SP 800-53 Rev. 5 control PM-12 asks the same of federal agencies and anyone who adopts the catalog.14

In practice a large organization's program has these parts:

  1. A charter approved by leadership. It says what the program is for (protecting people, data, money and operations), what it will and will not do, who is on the team, who sees what, and how employee privacy is protected. It is reviewed by the compliance officer and HR and, where employees are represented, discussed with the union.
  2. A cross-discipline team. HR, information security, IT, the compliance and privacy offices, physical security, and a manager from the business. Some programs add the employee assistance program or a behavioral expert for assessments. No single function decides alone; that protects employees from one department's bias and the organization from blind spots.
  3. A list of what matters most. The "crown jewels": trade secrets, source code, patient or customer databases, payment systems, plant controls, executive email. Monitoring focuses there, not on everyone equally.
  4. Technical sources. Logs of access to sensitive data, DLP alerts, cloud and email audit logs, badge records, and sometimes user and entity behavior analytics (UEBA), which flags activity that departs from a person's normal pattern, such as a bulk download the week after a resignation. These depend on the logging covered in lesson 10 and the data classification in lesson 7.
  5. Human sources. Managers and coworkers who report concerns through a known channel, and HR events such as resignations, terminations, disciplinary actions or a role change that removes access. CISA's fact sheet on HR's role stresses that HR often holds the earliest signals and must be part of the team.35
  6. An assessment process. Most alerts have innocent explanations. The team reviews context before anyone acts, documents the decision, and escalates only when the facts support it.
  7. A range of responses. Most cases end with a conversation, a reminder, a change in access or a referral for help with stress, debt or a workplace conflict. Some end in discipline or termination; a few involve law enforcement. CISA's guide devotes a chapter to managing a person of concern, including intervention, suspension and termination, and when to involve police.7
  8. Oversight and records. Case files are restricted, retention is limited, and the program reports aggregate results (not names) to leadership and the audit committee.
An insider risk program: four steps and one team 1. Define List the crown jewels. Write the charter and the limits on monitoring. 2. Detect and identify Access logs, DLP alerts, behavior analytics, HR events, reports from managers and staff. 3. Assess Review context first. Rule out innocent explanations. Document every decision. 4. Manage Conversation, access change, support, discipline; police only in rare cases. lessons feed back The team HR, security, IT compliance, privacy physical security business manager Most alerts have innocent explanations, and there is no reliable profile of an insider. Steps from CISA's Insider Threat Mitigation Guide (November 2020).
The four steps of an insider risk program, run by a cross-discipline team. Original graphic for this module; steps from CISA's Insider Threat Mitigation Guide, November 2020, and team roles from CISA's guide and 2026 team infographic.

There is no profile

One finding in CISA's guide deserves special attention from HR and managers: the 2020 edition's section on threat assessment is titled, in part, "No Useful Profile," and the September 2026 edition says "there is no demographic profile of an insider threat." There is no reliable demographic or personality profile of an insider, and programs that try to build one end up targeting people for who they are rather than what they do.7 Good programs look at behavior and access (a sudden bulk download, an attempt to reach systems outside one's job, a disabled security tool), at concerning statements or conduct reported by coworkers, and at stressors only as context, never as grounds by themselves. That approach is also what keeps a program on the right side of anti-discrimination law.

Monitoring employees' use of company systems is lawful in most U.S. workplaces, but not without limits. This section describes the main limits in general terms only; the details vary by state and change often, so your compliance officer and HR should check the current text in every state where you employ people before a program starts.

  • Federal wiretap and stored communications law. The Electronic Communications Privacy Act generally prohibits intercepting communications, but it has exceptions that employers rely on: consent of one party (which a signed, clear monitoring notice supplies), and an exception for the provider of the communication system acting to protect its rights or property. Listening to personal calls or reading personal accounts the employer does not run raises more risk than reviewing the company's own email system.36
  • State notice laws. New York's Civil Rights Law 52-c, in effect since May 2022, requires private employers that monitor employees' phone, email or internet use to give written notice at hiring, get an acknowledgment and post a notice; the attorney general can impose penalties of up to $500, $1,000 and $3,000 for first, second and later violations. Connecticut and Delaware have older notice laws, and other states have proposed similar ones.37
  • State privacy laws. California's privacy law (the CCPA as amended by the CPRA) has applied to employee and applicant data since Jan. 1, 2023, which brings notice, access and deletion rights to some employee monitoring data. Lesson 12 covers state privacy laws in detail (see lesson 12).
  • Labor law. Section 7 of the National Labor Relations Act protects employees who discuss pay and working conditions together, union or not. In October 2022 the NLRB's general counsel issued a memo (GC 23-02) urging a test that would have restricted intrusive electronic monitoring; on Feb. 14, 2025, the acting general counsel rescinded it.38 Monitoring that targets organizing remains unlawful, and in a unionized workplace new monitoring may need to be bargained.
  • Public employers. Government agencies, including Idaho state agencies, cities, counties and school districts, are bound by the Fourth Amendment. The Supreme Court has held that searches of employees' workspaces and devices are judged by reasonableness given the work-related purpose (O'Connor v. Ortega, 1987), and upheld a city's review of a police officer's text messages on a city pager as reasonable (City of Ontario v. Quon, 2010). Public employers also face public records laws, which can make monitoring records themselves requestable.39
  • Health and disability information. Information from an employee assistance program, medical leave or disability accommodation is protected by other laws and must not flow into an insider risk file without a lawful basis.
  • Proportionality. Even where the law allows broad monitoring, keystroke logging, webcam capture and personal-device monitoring are high-intrusion tools that damage trust. Use them only for defined high-risk roles or specific investigations, with approval, and say so in policy.

The topic is covered at small-business scale in module 73's Records Retention, Legal Holds and Workplace Privacy.

North Korean IT worker schemes

Since about 2020, thousands of IT workers dispatched by the North Korean government have used stolen or borrowed identities to get remote jobs at companies in the United States and elsewhere, mostly in software development and IT. Their wages go to the regime and its weapons programs. Some have also stolen source code and data, and since late 2024 some have threatened to leak it unless the employer pays. This is not a niche problem. Verizon's 2025 report cited the scheme in its discussion of privilege misuse, quoting the Justice Department's description of workers who stole "proprietary source code" and then demanded extortion payments.33

What the Justice Department cases show

The schemes depend on Americans who host company laptops at their homes ("laptop farms"), install remote-access software so the worker overseas appears to be logging in from the United States, and launder the pay. The Justice Department's DPRK RevGen: Domestic Enabler Initiative, a joint effort of its National Security Division and the FBI's Cyber and Counterintelligence Divisions, has gone after those enablers since 2024. A few of the public results:

  • Jan. 23, 2025: FBI PSA I-012325-PSA warned that North Korean IT workers had moved to data extortion, publicly releasing some victims' source code, copying code repositories to personal accounts, possibly harvesting credentials and session cookies, and using AI and face-swapping tools during video interviews.40
  • June 30, 2025: The Justice Department announced coordinated actions including searches of 29 known or suspected laptop farms across 16 states and seizure of about 200 computers. In one case alone, the scheme used the identities of more than 80 U.S. people to get jobs at more than 100 U.S. companies and caused at least $3 million in losses; a California defense contractor developing AI-powered equipment had export-controlled technical data accessed between January and April 2024.41
  • July 24, 2025: Christina Marie Chapman of Litchfield Park, Arizona, was sentenced to 102 months in prison for a scheme that generated more than $17 million for herself and North Korea. Workers using 68 stolen U.S. identities were placed at 309 U.S. businesses, including Fortune 500 companies, a top-five television network, an aerospace manufacturer and a car maker; more than 90 laptops were seized from her home. The workers also tried, mostly without success, to get jobs at two federal agencies. The U.S. attorney said: "Corporations failing to verify virtual employees pose a security risk for all."8
  • May 6, 2026: The Justice Department announced that two more facilitators, from Tennessee and New York, had each been sentenced to 18 months. Their schemes affected nearly 70 U.S. companies and generated more than $1.2 million for North Korea, and victim companies spent more than $1.5 million combined on remediation. The Justice Department said this was part of a series of announcements under the initiative in January 2025, June 2025, November 2025 and April 2026.42

Hiring controls that stop fake workers

The FBI's January 2025 PSA lists steps for employers, and they read like a checklist for HR, recruiters and IT working together:40

  1. Verify identity during interviewing, onboarding and throughout employment, for every remote worker.
  2. Cross-check HR records for applicants with matching résumé content or contact details, and watch for phone numbers (especially internet-based numbers) and emails reused across several résumés.
  3. Train HR staff, hiring managers and development teams on the scheme, especially on changes to address or payment method during onboarding.
  4. Confirm that staffing firms and contractors follow strong hiring practices, and audit them.
  5. Ask casual questions about where the applicant lives and studied, and look for inconsistencies.
  6. Do as much of hiring and onboarding in person as possible.

IT can add its own checks: ship laptops only to an address verified against identity documents, watch for remote-access tools and VPN or proxy use on corporate devices, flag a device whose network location does not match where the employee says they live, and require that payroll go to an account in the employee's verified name. The IC3 report's description of interviews where lip movements did not match the audio is another red flag interviewers can be told about.9

Hiring, background checks and separation of duties

Background checks

Background screening is a security control, but it is also one of the most regulated things an employer does. In general terms:

  • The Fair Credit Reporting Act (FCRA) applies whenever an employer uses a third-party screening company. The FTC's guidance lists the steps: a stand-alone written disclosure that a report may be obtained (not buried in the application), written authorization, a certification to the screening company, a pre-adverse action notice with a copy of the report and the "Summary of Your Rights" before rejecting someone because of it, and an adverse action notice afterward.43
  • The EEOC's 2012 enforcement guidance on arrest and conviction records says blanket exclusions can violate Title VII of the Civil Rights Act and asks employers to consider the nature and gravity of the offense, the time that has passed, and the nature of the job, with an individualized assessment.44
  • State and local "fair chance" laws limit when criminal history can be asked about and how it can be used. They vary widely; check the current text in each state and city where you hire.
  • Sector rules require more. Insured banks may not employ people convicted of certain crimes involving dishonesty, breach of trust or money laundering without the FDIC's consent under Section 19 of the Federal Deposit Insurance Act, and federally insured credit unions face a parallel rule in the Federal Credit Union Act; broker-dealers fingerprint many employees under SEC rules; health care organizations check federal exclusion lists; and cleared contractors run the federal investigation process.45

Screening is a snapshot. Most insider cases involve people who had clean records when hired. That is why rescreening for high-risk roles where the law allows it, and the ongoing controls below, matter more than a perfect hiring check.

Separation of duties and two-person rules

Separation of duties means no single person can complete a sensitive task alone. NIST SP 800-53 Rev. 5 control AC-5 asks organizations to identify duties that should be separated and to define access to support that.14 It is old advice from accounting, and it works against both fraud and mistakes:

  • The person who adds a vendor or changes its bank account is not the person who approves payments to it.
  • The developer who writes code does not approve its release to production; a second reviewer does.
  • The administrator who grants privileged access does not review the logs of their own activity.
  • The help desk agent who resets an executive's MFA needs a second confirmation (see the procedure above).
  • Bulk exports of customer or patient data need an approval and leave a record.
  • Emergency "break-glass" accounts require two people to open and trigger an alert when used.

Mandatory vacations for people in money-handling and privileged roles, long used by banks, serve the same purpose: some frauds only work if the person running them is always there to cover them up. The access rules behind all this (least privilege, privileged access management and access reviews) are in lesson 5.

Joiners, movers and leavers

The riskiest moments in an employee's life cycle are the edges: the first week, when accounts are being set up, and the last, when someone may be angry or taking files to a new job. Movers are the quiet problem; people who change roles often keep old access for years. For a large organization the people side of this is a handoff between HR and IT that must be automated wherever possible: HR's system of record triggers account creation, changes and removal, with same-day removal for involuntary departures and removal at the moment of notice for high-risk roles. Lesson 5 covers the technical side; module 73's lesson covers the offboarding sequence step by step.

Remote and hybrid work

The Twitter case began with remote workers taking calls about VPN problems at home. Remote and hybrid work is now permanent for much of the white-collar workforce, and it changes the human risks in specific ways: employees are alone when a convincing call arrives, cannot lean over to ask a coworker whether an email looks right, mix work and personal devices, and work in homes, cafés and hotels where others can see screens and hear calls.

NIST SP 800-46 Rev. 2, Guide to Enterprise Telework, Remote Access, and Bring Your Own Device (BYOD) Security (July 2016), is still the federal baseline. Its central assumption is that telework devices and the networks they use should be treated as untrusted, and that organizations should plan their remote access around that.46 In plain terms, for a large organization:

  • Company devices for company data, managed and encrypted, with EDR and automatic updates. Where personal devices are allowed, limit them to managed apps and web access with data kept in the company's cloud (see the BYOD policy in lesson 6).
  • Phishing-resistant sign-in for remote access, so a fake VPN page like the one in the Twitter case collects nothing usable.
  • A known way to verify IT. Publish the real service desk number on the intranet and badge, and tell employees that IT will never call and ask them to type credentials into a link, install software or read back a code.
  • Rules for the home office: lock the screen, keep work calls involving regulated data private, do not let family use work devices, shred or return printed sensitive documents, and use a privacy screen in public.
  • A channel for quick questions. A chat channel or hotline where remote staff can ask "is this real?" and get an answer in minutes replaces the coworker who used to sit nearby.
  • Travel guidance for executives and staff crossing borders with devices that hold sensitive data.
A woman sits at a desk with her laptop, focused on her work late at night in a low-light apartment.
A focused worker uses a laptop late at night in a low-light apartment. For a security program, this is the new front line: one person, alone, at the end of a long day, with no coworker nearby to ask whether a call from "IT" is real. Remote work rules work best when they give that person a fast way to check. Photo: Shixart1985, 2023-01-13. Wikimedia Commons, CC BY 2.0.

Acceptable use

An acceptable use policy (AUP) tells everyone with access what they may and may not do with the organization's systems and data, and what the organization may do in return, including monitoring. It is the policy most employees actually sign, so it should be short enough to read and plain enough to understand. A good AUP for a large organization covers:

  • Business use, with reasonable personal use allowed or not, stated clearly.
  • No expectation of privacy on company systems, with a plain description of what is monitored and why, to support the notice and consent the law requires.
  • Protecting credentials and MFA devices, and never sharing codes.
  • Handling data by classification (see lesson 7): no regulated data in personal email, personal cloud storage or unapproved apps.
  • Generative AI: which tools are approved, what data may never go into them, and how outputs must be checked.
  • Software and devices: no unapproved software, browser extensions, remote-access tools or USB storage.
  • Social media and public statements about the organization's systems.
  • Reporting: what to report, how, and the promise of no retaliation for good-faith reports.
  • Consequences, applied consistently.

Generative AI deserves its own attention. Verizon's 2025 report found that 15 percent of employees in its partner's data were routinely accessing generative AI systems on corporate devices (at least once every 15 days), and that many of them used personal email addresses as their sign-in, outside company control.23 Every paste of a contract, patient note or customer list into an unapproved tool is a disclosure the organization cannot take back. The answer that works is to provide an approved tool with contractual data protections, block or warn on unapproved ones, and say plainly in the AUP what data may never be entered.

Culture and reporting without blame

Every number in this lesson points to one conclusion: an organization's best human control is a workforce that reports fast, including reporting its own mistakes. That only happens when people believe they will be thanked rather than punished.

The idea comes from aviation and health care safety, where "just culture" programs distinguish honest error (console the person, fix the system), at-risk behavior (coach the person, remove the incentive for the shortcut) and reckless or malicious conduct (discipline). Applied to security:

  • An employee who clicks a phish and reports it within minutes has done the most valuable thing possible. Say so publicly, without naming them unless they want it.
  • An employee who hides a click for a week has turned a small problem into a large one. That is the behavior to address, not the click.
  • A help desk agent who follows the procedure and refuses an angry executive deserves public support from leadership.
  • Managers should hear about repeat problems in their teams as signals about workload, tools and process, not as names to discipline.
  • Reporting channels must be easy: a button in the email client, a short internal phone number, a chat channel, and an anonymous option for concerns about colleagues.

This is also where the board comes in. Directors set culture by what they ask about. A board that asks only for training completion rates gets compliance theater; a board that asks how fast the last real phishing campaign was reported and contained, and whether any executive asked for an exception to the payment rules, gets a security culture (see lesson 4).

FEMA public information officers seated in a classroom training session in Emmitsburg, Maryland, in January 2008.
FEMA public information officers in a classroom training session at the Emergency Management Institute in Emmitsburg, Maryland, January 16, 2008. Emergency management learned long ago that people perform under pressure the way they have practiced. Security training works the same way: short, realistic practice of the one action that matters (verify, then report) beats a long annual lecture. Photo: Bill Koplitz, 2008-01-16. Wikimedia Commons, Public domain.

A twelve-month people program

For a large organization starting over or rebuilding, the work fits into a year. The order matters: fix the processes criminals exploit most (help desk and payments) before spending more on general awareness.

  1. Months 1 to 3: close the big doors. Write and enforce the help desk verification procedure, including for any outsourced desk. Put the payment-change callback and dual-approval rule in writing, signed by the CEO and CFO. Deploy a one-click report button wired to the security team. Start phishing-resistant MFA for administrators, executives, finance and the help desk itself.
  2. Months 3 to 6: build the program. Charter the insider risk team and get the charter reviewed by HR and the compliance officer. Publish the updated acceptable use policy and monitoring notice, collecting acknowledgments where state law requires them. Launch role-based training tracks for finance, help desk, HR and recruiting, executives and developers. Add identity checks for remote hires, following the FBI's 2025 list.
  3. Months 6 to 9: measure and practice. Run announced phishing simulations rated with the NIST Phish Scale, focused on reporting. Run a tabletop exercise on a deepfake payment request and one on a departing employee taking data. Audit a sample of help desk resets and payment changes.
  4. Months 9 to 12: report and adjust. Give the board report rates, containment times, audit results and exceptions. Review the insider risk program's cases in aggregate for fairness. Update training to the year's real incidents. Rescreen high-risk roles where the law allows.

Each technical piece of this plan (MFA rollouts, DLP rules, report-button integrations) is a production change. Plan it, test it and have a way back.

A twelve-month people program Month 0 3 6 9 12 Close the big doors Help desk verification procedure, incl. vendors Payment callback and dual approval, signed One-click report button Phishing-resistant MFA: admins, executives, finance, help desk Build the program Insider risk charter and team Acceptable use policy and monitoring notice Role-based training tracks Identity checks for remote hires Measure, practice Announced simulations, difficulty rated Tabletops: deepfake payment, departing employee Audit resets and payment changes Report and adjust Board: report rates, containment time Fairness review of insider cases Training updated to real incidents Rescreen high-risk roles where lawful
A one-year order of work for the people side of a security program. Original graphic for this module; steps drawn from FBI PSA I-012325-PSA (Jan. 23, 2025), CISA's Insider Threat Mitigation Guide (2020), NIST SP 800-50 Rev. 1 (2024) and the NIST Phish Scale (2020).

Sample: Acceptable Use Policy

Sample only. Adapt it to your organization's risks, systems and laws, have qualified people review it, and do not treat it as a substitute for a risk assessment.

Follows: NIST SP 800-53 Rev. 5 controls PL-4 (Rules of Behavior), AC-5, AT-2 and PS-6; NIST SP 800-46 Rev. 2; FTC Safeguards Rule 16 CFR 314.4(e); HIPAA 45 CFR 164.308(a)(5); 23 NYCRR 500.14; New York Civil Rights Law 52-c (monitoring notice); FBI PSAs I-012325-PSA and I-121925-PSA.

1. Purpose and scope

This policy sets the rules for using [Organization]'s computers, phones, networks, accounts, cloud services, software and data ("systems"). It applies to every employee, officer, director, contractor, temporary worker, intern, volunteer and vendor employee who is given access to [Organization] systems ("users"), wherever they work, and to personal devices used for [Organization] work.

2. Business use

Systems are provided for [Organization] business. Limited personal use is [permitted / not permitted] if it does not interfere with work, does not violate law or policy, does not use significant resources, and does not involve [Organization] data. Personal use is not private (see section 9).

3. Accounts and sign-in

Users must: keep passwords and passkeys secret and unique to [Organization]; use the multifactor authentication method assigned to them; never share an account, a sign-in code or a security key; never approve a sign-in prompt they did not start; and report a lost device or security key to the service desk at [number] immediately. No legitimate [Organization] staff member will ask a user for a password or sign-in code.

4. Handling data

Users must handle data according to its classification under the [Data Classification and Retention Policy]. Users must not: send [Confidential] or [Restricted] data to personal email or personal cloud storage; store it on unapproved devices or removable media; share it outside [Organization] except through approved tools and with authorization; or access records they do not need for their job. Users must double-check recipients before sending [Confidential] or [Restricted] data and report any misdirected message at once.

5. Verification rules

Users must not change payment details, send money, release payroll or tax records, reset credentials, install software or grant access based on a single message, call or video meeting, no matter who appears to be asking. Such requests must be confirmed through a separate channel already on file, under the [Payment Verification Procedure] and [Service Desk Identity Verification Procedure]. No officer of [Organization] may waive this section.

6. Software, devices and AI tools

Only software, browser extensions, remote-access tools and cloud services approved by [IT/Security] may be used. Users must not disable security tools, connect unapproved storage devices, or connect unapproved devices to [Organization] networks. Only the generative AI tools listed at [location] may be used for [Organization] work, and [Confidential] or [Restricted] data, including [patient / customer / client / student] information, source code and credentials, must never be entered into any other AI tool. Users are responsible for checking AI output before relying on it.

7. Remote and mobile work

Remote users must use [Organization]-managed devices [or enrolled personal devices under the BYOD Policy], connect through [approved remote access], lock screens when away, keep work calls involving [Confidential] data private, prevent family members or others from using work devices, and secure or return printed sensitive documents. Users traveling internationally must follow the [Travel Device Procedure].

8. Prohibited conduct

Users must not use systems to: break the law; harass or discriminate; access, download or share sexually explicit material; infringe copyrights or licenses; attempt to get around security controls or access systems they are not authorized to use; or conduct personal business for profit. Security testing of [Organization] systems requires written authorization from [CISO].

9. Monitoring and no expectation of privacy

[Organization] monitors its systems to protect people, data and operations and to comply with law. Users' communications and internet use on [Organization] systems may be subject to monitoring at any and all times and by any lawful means, including email, messaging, file access, web traffic, sign-in records and data transfers. [Organization] does not [record keystrokes / activate cameras or microphones / monitor personal devices beyond the managed work profile] except [describe approved, limited circumstances, such as an authorized investigation approved by HR and the compliance officer]. Monitoring data is restricted to authorized staff, kept for [period], and used only for the purposes in this policy.

10. Reporting, without blame

Users must report suspected phishing, lost devices, mistakes and security concerns promptly using [report button / number / channel]. Users who report their own good-faith mistakes promptly will not be disciplined for the mistake. [Organization] will not retaliate against anyone for a good-faith report, including a concern about a colleague, which may be made anonymously at [channel].

11. Exceptions and enforcement

Exceptions require written approval from [CISO] and [data owner], with an expiration date. Violations may result in loss of access and discipline up to termination of employment or contract, and may be reported to law enforcement where required. Discipline will be applied consistently, taking into account intent and prompt reporting.

12. Acknowledgment and review

Users must acknowledge this policy [in writing or electronically] at hire or before access is granted, and [annually]. [Organization] will post this notice at [location]. [Owner] reviews this policy at least annually. Approved by [title] on [date]. Version [number].

Pictures, Numbers and Infographics

The key numbers in this lesson, each with its source and year.

NumberWhat it measuresSource
60%Share of breaches with a human element (61% the year before)Verizon DBIR 20251
72%Share of end-user error actions that were misdeliveryVerizon DBIR 20251
About 21% vs. 5%Simulated phishing report rate, recently trained vs. notVerizon DBIR 202518
15%Employees routinely using generative AI on corporate devicesVerizon DBIR 202523
1,008,597; $20.877 billionAll complaints and reported losses, 2025FBI IC3 2025 report2
191,561Phishing and spoofing complaints, 2025FBI IC3 2025 report2
$3,046,598,558Business email compromise losses, 2025FBI IC3 2025 report2
22,364; $893,346,472AI-related complaints and losses, 2025FBI IC3 2025 report9
4,479; $88,725,284Idaho complaints and reported losses, 2025FBI IC3 2025 report2
32.10%Employees who clicked at least one of eight simulated phishes (14,000+ staff, 15 months)Lain, Kostiainen and Capkun, 20223
About 2%Reduction in clicking from embedded training (19,500+ staff)Ho and colleagues, 20254
30.8% vs. 1.82%Click rates on a vacation-policy lure vs. a password-update lureHo and colleagues, 202517
About 24 hours; 130 accounts; about $118,000Twitter hack of July 15, 2020: duration, accounts compromised, bitcoin stolenNYDFS report, Oct. 20205
HK$200 million (about US$25.6 million); 15 transfersArup deepfake video conference fraud, January 2024South China Morning Post, May 20246
$380 millionDamages Clorox seeks from its former help desk provider (pending)Clorox complaint, July 202526
$180 million to $400 millionCoinbase's estimated cost of the bribed-insider breachCoinbase Form 8-K, May 14, 202532
309 companies; 68 identities; 102 monthsChapman laptop farm case: companies defrauded, stolen identities, prison sentenceJustice Department, July 24, 20258
29 laptop farms; 16 statesSearches in the June 2025 coordinated actionsJustice Department, June 30, 202541
$500; $1,000; $3,000Maximum New York penalties for first, second and later monitoring notice violationsN.Y. Civil Rights Law 52-c37
At least once every 12 monthsSecurity awareness training frequency in NYDFS 500.14 and PCI DSS 12.623 NYCRR 500.14; PCI DSS v4.0.11215
People-driven crime reported to the FBI in 2025 (complaints) Drawn to scale, 0 to 200,000. Categories can overlap; AI-related is a cross-cutting tag. Phishing and spoofing 191,561 Personal data breach 67,456 Tech support scams 47,794 Government impersonation 32,424 Employment scams 24,688 Mentioned AI 22,364 ($893.3 million lost) 0 100,000 200,000 Source: FBI Internet Crime Report 2025 (released April 2026).
Complaint counts for the IC3 categories that most often begin with a person being deceived. Original graphic for this module; data from the FBI Internet Crime Report 2025, released April 2026.

The other infographics in this lesson: the research comparison in What the research says, the help desk flow in The help desk is the front door, the insider risk cycle in Building the program, and the one-year plan in A twelve-month people program.

Wide panoramic view of a home office with a desk by a window and bookshelves on both sides.
A panoramic view of a home office in April 2020, photographed by Bill Smith of Virginia Beach, where the owner planned to move his laptop to the desk by the window to teach classes remotely instead of meeting students on campus. Millions of workers made the same move that spring, which is when the Twitter attackers found remote employees ready to believe a call about VPN trouble. Photo: Bill Smith from Virginia Beach, 2020-04-22. Wikimedia Commons, CC BY 2.0.

Common mistakes

  • Trusting a voice that says it is IT. Twitter's remote employees in 2020 believed a caller claiming to be the help desk and entered credentials and codes into a fake VPN page. New York regulators found the attack ran from first calls to takeover of 130 accounts in about a day.5 Fix: publish the real IT number, tell staff IT never asks for credentials or codes, and use phishing-resistant MFA.
  • A help desk that can be talked into resets. The CISA and FBI Scattered Spider advisory describes attackers posing as employees to get passwords reset and MFA moved to their own devices, including through contracted help desks.24 Clorox's pending suit alleges exactly that happened to it, and seeks $380 million; Cognizant denies responsibility.26 Fix: the verification procedure in this lesson, written into the vendor contract and audited.
  • Letting a video call authorize a payment. Arup's employee saw and heard what looked like the CFO and colleagues and sent HK$200 million in 15 transfers.6 Fix: no payment on a single channel, including video, and no executive waiver.
  • Giving support agents bulk access to customer data. Coinbase's 2025 breach came from bribed contractor support agents who could see ID images, balances and contact details; the company estimated $180 million to $400 million in costs.32 Fix: minimum data per ticket, alerts on unusual lookups, and contractor staff inside the insider risk program.
  • Hiring remote workers without real identity checks. More than 300 U.S. companies, including Fortune 500 firms, unknowingly employed North Korean workers placed through one Arizona laptop farm.8 Fix: the FBI's 2025 hiring checklist, in-person steps where possible, and device location checks.40
  • Counting completions and calling it a program. UC San Diego Health's experiment found no meaningful link between annual training completion and falling for phishing.4 Fix: measure reporting speed, real catches and process audits.
  • Punishing clickers. Lain and colleagues found that the most valuable thing employees did was report, and that reports arrive within minutes when people are encouraged.3 A culture of punishment produces hidden clicks. Fix: thank reporters, address hiding rather than clicking.
  • Monitoring without notice. New York requires written notice, acknowledgment and a posted notice for electronic monitoring, with penalties for each violation.37 Fix: a plain monitoring section in the acceptable use policy and acknowledgments collected where required.

Questions people ask

Is annual training required even if research says it barely works?

Often yes. NYDFS 500.14 requires training at least annually, PCI DSS requires it at hire and every 12 months, and HIPAA and the FTC Safeguards Rule require a training program.121110 The research does not make those requirements go away; it tells you to meet them efficiently and spend the real effort on reporting, role-based training and process controls.

Should we run phishing simulations at all?

Yes, if they are announced in policy, rated for difficulty, aimed at practicing reporting, and free of cruel lures. They also show where your real defenses fail. Do not rely on embedded "you clicked" lessons as the main tool; two large studies found them weak or counterproductive.34

Can we fire someone for clicking a phishing link?

Legally, in most at-will workplaces, often yes. As security practice, it is usually a mistake for a single honest click, because it teaches everyone else to hide clicks. Reserve discipline for hiding incidents, deliberately bypassing controls or repeated refusal to follow verification rules, and apply it consistently.

How can staff tell a deepfake from a real person?

Reliably, they often cannot, and the FBI says synthetic content is getting harder to detect.9 Do not make staff judges of authenticity. Make the rule that certain actions never happen on one channel, and have them confirm through a known number or a pre-agreed second channel.

Does an organization our size need a formal insider risk program?

Cleared defense contractors must have one under 32 CFR 117.7.16 For other large organizations, NYDFS requires monitoring of authorized users' activity, and the NIST catalog includes an insider threat program control.1214 Even where not required, any organization with crown-jewel data and hundreds of employees benefits from at least a chartered team, defined data to watch and a fair process.

What should a public agency do differently?

Public employers face the Fourth Amendment's reasonableness test for searches, public records laws that can reach monitoring data, and, in Idaho, the state's public records statute in Idaho Code Title 74. Write the purpose and scope of monitoring into policy, keep it work-related, and plan for how records requests for monitoring data will be handled.3947

Are contractors and vendors covered by these training rules?

Usually. HIPAA covers the whole workforce, NYDFS says all personnel, and PCI DSS covers personnel with access. Contracts should require vendors' staff with access to complete equivalent training and follow your verification rules (see lesson 8).1112

What happens if an employee emails personal data to the wrong person?

Treat it as a possible breach: retrieve or get confirmation of deletion, document what was sent, and run the analysis under state breach laws and any sector rule. Idaho's law requires notice only when an investigation finds misuse has occurred or is reasonably likely, but other states and HIPAA use different tests. Lesson 11 walks through it (see lesson 11).48

Hard questions

Is security awareness training worth the money?

The case for cutting it back: The two largest independent field studies found that common training formats did little to reduce clicking, and the authors of the 2025 study concluded current programs are unlikely to offer significant practical value against phishing. Money spent on phishing-resistant MFA, email filtering and payment controls reduces risk more reliably.43

The case for keeping it: Rules require it. Training measurably raises reporting, which Verizon found quadrupled after recent training, and reporting catches campaigns early. Some attacks, such as deepfake payment requests and help desk pretexts, have no purely technical fix.18

What the evidence shows: Both are partly right. Training that aims to stop every click fails; training that aims to get fast reports and to teach a few high-risk roles specific verification rules works. Fund technical controls first, then fund training for reporting and roles.

How much employee monitoring is too much?

The case for broad monitoring: Insiders cause real losses, as Coinbase's bribed agents and North Korean workers show. Without logs and behavior analytics an organization cannot detect a bulk download or a laptop being controlled from abroad, and regulators such as NYDFS require monitoring of user activity.3212

The case for restraint: Keystroke logging, screen capture and webcam monitoring treat every employee as a suspect, damage trust and morale, collect sensitive personal data that then must be protected, and can chill lawful activity such as discussing pay. CISA's own guide warns there is no useful profile of an insider, which cuts against broad behavioral surveillance.7

What the evidence shows: The defensible middle is monitoring focused on crown-jewel data and privileged access, disclosed in policy, reviewed by a cross-discipline team, with high-intrusion tools reserved for approved investigations. That catches the cases in the public record without monitoring everyone's every keystroke.

Should phishing tests use the most realistic lures, such as fake bonuses?

The case for realism: Criminals use exactly those lures. The UC San Diego study found a workplace-policy lure fooled 30.8 percent of employees.17 Testing with soft lures gives leaders false comfort.

The case against: Lures about pay, layoffs or health exploit employees' real anxieties, have caused public backlash when organizations used them, and erode the trust that makes people willing to report.

What the evidence shows: Difficulty can be raised without cruelty: use realistic invoices, shared-document notices and policy updates, rate them with the NIST Phish Scale, and warn staff about the cruel lures in training rather than springing them.19

Does remote work make organizations less secure?

The case that it does: The Twitter attack exploited remote employees' routine VPN trouble, North Korean schemes depend on fully remote hiring, and remote workers lack the coworker who would say "that call sounds odd."541

The case that it need not: Office networks are breached too, and the strongest controls (phishing-resistant MFA, managed devices, identity verification, verification rules for payments) work the same at home. NIST's telework guidance assumes untrusted networks and shows how to design for them.46

What the evidence shows: Remote work shifts risk to identity and verification. Organizations that invest in those controls and in fast ways for staff to check suspicious contacts can run remote work safely; those that simply extended the office VPN to homes inherited the risk.

For reporters and advocates

What public records exist.

  • SEC filings. Public companies disclose material cyber incidents on Form 8-K Item 1.05 and describe their programs, including training and insider risk, in Form 10-K Item 1C. Coinbase's May 14, 2025 8-K and MGM's Oct. 5, 2023 8-K are examples. Search EDGAR's full-text search for "social engineering," "insider" or "help desk."4932
  • Justice Department releases and court dockets. North Korean IT worker prosecutions, insider data theft cases and extortion cases are announced at justice.gov and filed on PACER and CourtListener. The releases often name the number of victim companies but not the companies; the indictments sometimes describe them.41
  • Civil suits between companies and their vendors. Clorox v. Cognizant shows how help desk procedures become evidence. State court dockets carry these cases.27
  • Regulators' reports. The NYDFS Twitter report is a model of what a regulator can publish.5 HHS OCR resolution agreements and its breach portal list health care breaches, including many caused by insiders and misdirected records.50
  • State attorney general breach lists. Maine's list, among others, shows the number of people notified and often the cause, including "insider wrongdoing" and "inadvertent disclosure."51
  • FBI and CISA alerts. The IC3 annual report (with state tables; Idaho had 4,479 complaints and $88,725,284 in reported losses in 2025) and the PSAs cited here are public.2
  • Idaho public records. Under Idaho Code Title 74, reporters can request a state agency's, city's, county's or school district's acceptable use and monitoring policies, security awareness training contracts and completion summaries, and help desk verification procedures, though specific security details may be withheld under the statute's exemptions.47

Questions to ask an organization. How does your help desk verify a caller before resetting a password or MFA, and is that rule in your outsourced service desk's contract? Has an executive ever asked for an exception to payment verification? What share of employees reported the last real phishing campaign, and how fast was it removed? Do you have an insider risk program, who sits on it, and what does it monitor? How do you verify the identity of remote hires? What do your employees sign about monitoring, and what does it say?

The Idaho angle. Idaho public officials are within the groups the FBI's 2025 impersonation warnings describe, and Idaho's state agencies, counties, cities, school districts and hospitals all hire remote IT staff and contractors. Ask the Idaho Technology Authority and large local governments whether they have adopted help desk verification and remote-hire identity checks, and ask hospitals and credit unions how they train staff on deepfake payment and identity requests.

Where to go next

Still being checked (4 items)

These points could not be fully confirmed against a primary source when this lesson was written and fact checked in November 2026. The text above states each one as uncertain.

  • Clorox v. Cognizant: after the March 25, 2026 ruling (Law360), a May 12, 2026 law firm commentary (Hunton) still described the case as ongoing; no trial date, settlement or later ruling could be found as of Nov. 14, 2026.
  • CISA's January 2026 infographic on insider threat management teams is described from WaterISAC (Feb. 5, 2026) and ExecutiveGov (Jan. 29, 2026) summaries; the CISA page itself did not load on Nov. 14, 2026.
  • PCI DSS v4.0.1 requirement numbers (12.6.3.1 and 12.6.3.2) and their March 31, 2025 effective date match several assessor summaries checked Nov. 14, 2026; the copyrighted standard itself was not reopened.
  • State fair-chance hiring laws and any Idaho-specific rules on criminal history in private hiring were not checked state by state.

Last verified November 14, 2026. General education, not legal, compliance, audit or other professional advice; rule versions and dates are stated in the text, and the sources used are listed in the Notes and Sources below. Check the regulator's current text before relying on any point.

Notes

  1. Verizon, 2025 Data Breach Investigations Report, April 23, 2025, pp. 20 to 22 (human element in 60 percent of breaches versus 61 percent the prior year; internal actor breaches, misdelivery 72 percent of end-user action varieties). Local copy saved in the module downloads (read Nov. 13, 2026). Read it online. ↩
  2. Federal Bureau of Investigation, Internet Crime Complaint Center, Internet Crime Report 2025 (released April 2026): 1,008,597 complaints, $20.877 billion in losses; complaint and loss tables by crime type; state tables (Idaho 4,479 complaints, $88,725,284) (read Nov. 13, 2026). Read it online. ↩
  3. Daniele Lain, Kari Kostiainen and Srdjan Capkun, "Phishing in Organizations: Findings from a Large-Scale and Long-Term Study," 2022 IEEE Symposium on Security and Privacy, pp. 842 to 859 (preprint arXiv:2112.07498) (read Nov. 13, 2026). Read it online. ↩
  4. Grant Ho, Ariana Mirian, Elisa Luo, Khang Tong, Euyhyun Lee, Lin Liu, Christopher A. Longhurst, Christian Dameff, Stefan Savage and Geoffrey M. Voelker, "Understanding the Efficacy of Phishing Training in Practice," 2025 IEEE Symposium on Security and Privacy (May 2025) (read Nov. 13, 2026). Read it online. ↩
  5. New York State Department of Financial Services, Twitter Investigation Report (report on Twitter's July 15, 2020 cybersecurity incident), October 2020 (read Nov. 13, 2026). Read it online. ↩
  6. South China Morning Post, "UK multinational Arup confirmed as victim of HK$200 million deepfake scam that used digital version of CFO to dupe staff," May 17, 2024 (15 transfers to five accounts; Arup statement that internal systems were not compromised) (read Nov. 13, 2026). Read it online. ↩
  7. Cybersecurity and Infrastructure Security Agency, Insider Threat Mitigation Guide, November 2020, and updated edition, September 2026 (the 2026 edition keeps the four-part structure and says "there is no demographic profile of an insider threat"; four steps: defining, detecting and identifying, assessing and managing the threat; "Profiles: No Useful Profile in Threat Assessment"; managing insider threats chapter) (read Nov. 13, 2026). Read it online. ↩
  8. U.S. Department of Justice, Office of Public Affairs, "Arizona Woman Sentenced for $17M Information Technology Worker Fraud Scheme that Generated Revenue for North Korea," July 24, 2025 (read Nov. 13, 2026). Read it online. ↩
  9. FBI Internet Crime Complaint Center, Internet Crime Report 2025, "Artificial Intelligence (AI) Used in Cybercrime," pp. 39 to 40 (22,364 AI-related complaints; $893,346,472 in losses; more than $30 million in AI-involved BEC; almost $13 million in AI-involved employment scams) (read Nov. 13, 2026). Read it online. ↩
  10. Federal Trade Commission, Standards for Safeguarding Customer Information, 16 CFR 314.4(e), as amended (amendments effective June 9, 2023; notification amendment effective May 13, 2024) (read Nov. 13, 2026). Read it online. ↩
  11. HIPAA Security Rule, Administrative safeguards, 45 CFR 164.308(a)(5) (security awareness and training) (read Nov. 13, 2026). Read it online. ↩
  12. New York Department of Financial Services, Cybersecurity Requirements for Financial Services Companies, 23 NYCRR 500.14 (Monitoring and training), as amended Nov. 1, 2023, text via Cornell Legal Information Institute (read Nov. 13, 2026). Read it online. ↩
  13. NIST, SP 800-171 Rev. 3, Protecting Controlled Unclassified Information in Nonfederal Systems and Organizations, May 2024, requirements 03.02.01 (Literacy Training and Awareness) and 03.02.02 (Role-Based Training) (read Nov. 13, 2026). Read it online. ↩
  14. NIST, SP 800-53 Rev. 5, Security and Privacy Controls for Information Systems and Organizations (current release on CSRC), controls AC-5 (Separation of Duties), AT-2 with enhancements (2) Insider Threat and (3) Social Engineering and Mining, AT-3, PL-4, PS-3, PS-6 and PM-12 (Insider Threat Program) (read Nov. 13, 2026). Read it online. ↩
  15. PCI Security Standards Council, PCI DSS v4.0.1, June 2024, Requirement 12.6 (security awareness program), described and not quoted; requirements 12.6.3.1 and 12.6.3.2 became mandatory March 31, 2025. The standard is available free after accepting the council's license in its document library (read Nov. 13, 2026). Read it online. ↩
  16. 32 CFR 117.7, Procedures (National Industrial Security Program Operating Manual), paragraphs (b)(4) and (d) (insider threat program and Insider Threat Program Senior Official, consistent with Executive Order 13587) (read Nov. 13, 2026). Read it online. ↩
  17. UC San Diego Today, "Cybersecurity Training Programs Don't Prevent Employees from Falling for Phishing Scams," Sept. 17, 2025 (study presented at the IEEE Symposium on Security and Privacy, May 2025, and at Black Hat, August 2025) (read Nov. 13, 2026). Read it online. ↩
  18. Verizon, 2025 Data Breach Investigations Report, pp. 47 to 49 (phishing simulation click rate distribution, about 1.5 percent median persistent clickers; report rate about 21 percent with training in the prior 30 days versus a base rate of 5 percent) (read Nov. 13, 2026). Read it online. ↩
  19. Michelle Steves, Kristen Greene and Mary Theofanos, "Categorizing human phishing difficulty: a Phish Scale," Journal of Cybersecurity 6, no. 1 (2020), doi:10.1093/cybsec/tyaa009, published online Sept. 14, 2020; NIST news release "The Phish Scale: NIST-Developed Method Helps IT Staff See Why Users Click on Fraudulent Emails," Sept. 17, 2020 (read Nov. 13, 2026). Read it online. ↩
  20. CISA, "Implementing Phishing-Resistant MFA" fact sheet, October 2022 (read Nov. 13, 2026). Read it online. ↩
  21. NIST, SP 800-50 Rev. 1, Building a Cybersecurity and Privacy Learning Program, September 2024 (supersedes SP 800-50 of 2003 and SP 800-16 of 1998; life cycle of plan and strategy, analysis and design, development and implementation, and assessment and improvement, with content built on the ADDIE model) (read Nov. 13, 2026). Read it online. ↩
  22. CISA, NSA, FBI and MS-ISAC, Phishing Guidance: Stopping the Attack Cycle at Phase One, October 2023 (read Nov. 13, 2026). Read it online. ↩
  23. Verizon, 2025 Data Breach Investigations Report, p. 13 and pp. 24 to 25 (synthetically generated text in malicious emails doubled over two years per a data partner; 15 percent of employees routinely accessing generative AI on corporate devices) and the Social Engineering pattern section (read Nov. 13, 2026). Read it online. ↩
  24. CISA, FBI and partners, Joint Cybersecurity Advisory AA23-320A, "Scattered Spider," first published Nov. 16, 2023, updated July 29, 2025 (TTPs through June 2025; help desk impersonation; DragonForce ransomware; mitigations) (read Nov. 13, 2026). Read it online. ↩
  25. MGM Resorts International, Current Report on Form 8-K, filed Oct. 5, 2023 (estimated negative impact of about $100 million on third-quarter adjusted property EBITDAR and less than $10 million in one-time expenses), accession no. 0001193125-23-251667, available on EDGAR (read Nov. 13, 2026). Read it online. ↩
  26. The Register, "$380M lawsuit claims intruder got Clorox's passwords from Cognizant simply by asking," July 23, 2025 (summarizing Clorox's complaint filed in California state court and Cognizant's response) (read Nov. 13, 2026). Read it online. ↩
  27. Law360, "Cognizant Must Face Clorox's $380M Suit Over Login Sharing," March 25, 2026 (judge dismissed intentional misrepresentation claim, let most of the suit proceed) (read Nov. 13, 2026). Read it online. ↩
  28. Financial Crimes Enforcement Network, Alert FIN-2024-Alert004, "FinCEN Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions," Nov. 13, 2024 (read Nov. 13, 2026). Read it online. ↩
  29. FBI, Public Service Announcement I-051525-PSA, "Senior US Officials Impersonated in Malicious Messaging Campaign," May 15, 2025 (read Nov. 13, 2026). Read it online. ↩
  30. FBI, Public Service Announcement I-121925-PSA, "Senior U.S. Officials Continue to be Impersonated in Malicious Messaging Campaign," Dec. 19, 2025 (read Nov. 13, 2026). Read it online. ↩
  31. CISA, "Defining Insider Threats" (definitions of insider and insider threat; negligent, accidental, intentional, collusive and third-party types) (read Nov. 13, 2026). Read it online. ↩
  32. Coinbase Global, Inc., Current Report on Form 8-K, May 14, 2025, and Coinbase, "Protecting Our Customers: Standing Up to Extortionists," May 2025 (bribed overseas support contractors; $20 million extortion demand refused; estimated $180 million to $400 million in remediation and reimbursement) (read Nov. 13, 2026). Read it online. ↩
  33. Verizon, 2025 Data Breach Investigations Report, Privilege Misuse pattern, pp. 62 to 63 (increase in Partner actors; convergence of end users and system administrators; North Korean workers quoted from Justice Department indictment) (read Nov. 13, 2026). Read it online. ↩
  34. WaterISAC, "TLP:CLEAR CISA Urges Critical Infrastructure Organizations to Take Action Against Insider Threats and Publishes New Product," Feb. 5, 2026 (describing CISA's infographic "Assembling a Multi-Disciplinary Insider Threat Management Team," released late January 2026) (read Nov. 13, 2026). Read it online. ↩
  35. CISA, "Human Resources' Role in Preventing Insider Threats" fact sheet (read Nov. 13, 2026). Read it online. ↩
  36. Electronic Communications Privacy Act, 18 U.S.C. 2511(2)(a)(i) and (2)(d) (provider and consent exceptions), text via Cornell Legal Information Institute (read Nov. 13, 2026). Read it online. ↩
  37. New York Civil Rights Law 52-c (electronic monitoring notice; effective May 7, 2022; civil penalties of up to $500, $1,000 and $3,000) (read Nov. 13, 2026). Read it online. ↩
  38. Littler, "Acting NLRB General Counsel Rescinds Controversial Memoranda," Feb. 2025 (GC 25-05, issued Feb. 14, 2025 by Acting General Counsel William Cowen, rescinding GC 23-02 on electronic monitoring and algorithmic management) (read Nov. 13, 2026). Read it online. ↩
  39. City of Ontario v. Quon, 560 U.S. 746 (2010), and O'Connor v. Ortega, 480 U.S. 709 (1987) (read Nov. 13, 2026). Read it online. ↩
  40. FBI, Public Service Announcement I-012325-PSA, "North Korean IT Workers Conducting Data Extortion," Jan. 23, 2025 (read Nov. 13, 2026). Read it online. ↩
  41. U.S. Department of Justice, Office of Public Affairs, "Justice Department Announces Coordinated, Nationwide Actions to Combat North Korean Remote Information Technology Workers' Illicit Revenue Generation Schemes," June 30, 2025 (updated July 3, 2025) (read Nov. 13, 2026). Read it online. ↩
  42. U.S. Department of Justice, Office of Public Affairs, "Two U.S. Nationals Sentenced for Facilitating Fraudulent Remote Information Technology Worker Schemes," May 6, 2026 (read Nov. 13, 2026). Read it online. ↩
  43. Federal Trade Commission, "Using Consumer Reports: What Employers Need to Know" (business guidance, October 2016, last modified Aug. 4, 2020) (read Nov. 13, 2026). Read it online. ↩
  44. U.S. Equal Employment Opportunity Commission, Enforcement Guidance on the Consideration of Arrest and Conviction Records in Employment Decisions under Title VII, April 25, 2012 (read Nov. 13, 2026). Read it online. ↩
  45. Federal Deposit Insurance Act Section 19, 12 U.S.C. 1829 (consent required to employ persons convicted of crimes of dishonesty, breach of trust or money laundering), text via Cornell Legal Information Institute; for federally insured credit unions, Federal Credit Union Act Section 205(d), 12 U.S.C. 1785(d); see also SEC Rule 17f-2 (fingerprinting of securities industry personnel) (read Nov. 13, 2026). Read it online. ↩
  46. NIST, SP 800-46 Rev. 2, Guide to Enterprise Telework, Remote Access, and Bring Your Own Device (BYOD) Security, July 2016 (read Nov. 13, 2026). Read it online. ↩
  47. Idaho Code Title 74, Chapter 1 (Public Records Act), including exemptions for certain security records (read Nov. 13, 2026). Read it online. ↩
  48. Idaho Code 28-51-105 (disclosure of breach of security of computerized personal information), as amended through the 2026 session (read Nov. 13, 2026). Read it online. ↩
  49. U.S. Securities and Exchange Commission, EDGAR full-text search (Form 8-K Item 1.05 and Form 10-K Item 1C filings) (read Nov. 13, 2026). Read it online. ↩
  50. U.S. Department of Health and Human Services, Office for Civil Rights, Breach Portal: Notice to the Secretary of HHS Breach of Unsecured Protected Health Information (read Nov. 13, 2026). Read it online. ↩
  51. Office of the Maine Attorney General, Data Breach Notifications list (read Nov. 13, 2026). Read it online. ↩

Check yourself

1. In the Verizon 2025 Data Breach Investigations Report, what share of breaches involved a human element?

2. What did Lain, Kostiainen and Capkun (IEEE Symposium on Security and Privacy, 2022) find about embedded training shown to people who click a simulated phish?

3. In Verizon's 2025 analysis of phishing simulation data, what happened to the report rate among employees trained in the previous 30 days?

4. According to the July 29, 2025 update of the CISA and FBI Scattered Spider advisory, how did the actors use help desks?

5. In the Arup case reported in 2024, what convinced the Hong Kong employee to make 15 transfers totaling HK$200 million?

6. Which of these is the strongest defense against a deepfake request to send an urgent payment?

7. What are the four steps of CISA's Insider Threat Mitigation Guide (first published November 2020, updated September 2026)?

8. Under 32 CFR 117.7, which organizations must have an insider threat program led by a named senior official?

9. Which step is on the FBI's January 2025 list for employers worried about North Korean IT workers?

10. What does New York Civil Rights Law 52-c require of private employers that monitor employees' email, phone or internet use?

Terms in this lesson

Social engineering
Manipulating a person into doing something that helps an attacker, such as revealing a password, approving a sign-in, resetting an account or sending money.
Pretexting
Inventing a believable story (for example, a VPN problem or an urgent confidential deal) to get someone to act.
Vishing
Voice phishing: social engineering by phone call or voice message, now sometimes using AI-cloned voices.
Deepfake
Synthetic audio, image or video created with artificial intelligence to imitate a real person.
Embedded training
A short lesson shown automatically to someone who clicks a simulated phishing email.
Phish Scale
A NIST method (2020) for rating how hard a simulated phishing email is to detect, so click rates can be compared fairly.
Report rate
The share of recipients who report a phishing email to the security team; a better measure of program health than click rate.
Insider threat
In CISA's words, the threat that an insider will use their authorized access, wittingly or unwittingly, to do harm.
Insider risk program
A chartered, cross-discipline program that identifies, assesses and manages risks from people with authorized access.
User and entity behavior analytics (UEBA)
Tools that flag activity departing from a person's or system's normal pattern, such as a bulk download after a resignation.
Separation of duties
Dividing a sensitive task so no single person can complete it alone, as in NIST SP 800-53 control AC-5.
Laptop farm
A home where company laptops shipped to a fake or borrowed identity are kept online so overseas workers appear to work from the United States.
Acceptable use policy (AUP)
The policy that sets what users may and may not do with an organization's systems and data, including monitoring notice.
Just culture
A safety approach that separates honest error, at-risk behavior and reckless conduct, consoling, coaching or disciplining accordingly.

Full glossary

Sources