AI Ethics, Disclosure and Privacy Rules

The ethics of AI come down to old rules applied to a new tool. Don't spread what hasn't been verified. Don't invent things. Protect the people who trusted you with information. Tell your audience how the work was done when it matters to them. What's new is how easy the tool makes it to break each rule without noticing.

Newsrooms wrote the most detailed versions of these rules, and they are worth knowing even outside journalism. They are the yardstick for judging AI use in local news coverage, and they apply just as well to a campaign flyer, an advocacy group's report or a county's public notice.

AP's standards

The Associated Press published its generative AI standards on August 16, 2023, and they became the reference point for much of the industry. The core rules, as AP described them and as widely reported:

• AI output is unvetted source material. It goes through the same editorial judgment and sourcing standards as anything else.
• Staff may experiment with tools like ChatGPT with caution, but not to create publishable content.
• No generative AI to add or remove elements from photos, video or audio. The exception: AI-generated images that are themselves the subject of a story, clearly labeled.
• Normal skepticism applies to suspect material: trace its source, run reverse image searches, check for similar reports from trusted outlets.
• No confidential or sensitive information goes into AI tools.
• Journalists stay accountable for accuracy and fairness. AI is not a replacement for them.

AP updates its guidance, so its standards page is the place to confirm any detail.

The 2026 SPJ Code of Ethics

The Society of Professional Journalists revised its Code of Ethics on October 4, 2026, at its national convention in Columbus, Ohio, the first revision since 2014. The code never names AI; the committee's explainer says a specific term would date it. Instead it adds lines that cover any tool. Among them:

The committee's explainer says it followed AP's approach: treat information from generative AI like any other source, which means verifying it. The code has no standalone AI disclosure rule; the transparency lines above do that work.

Never fabricate

This one has no gray zone. These are fabrication whether a human or a model produced them:

• A quote from a person who doesn't exist, including "composite" residents
• Words in quotation marks that a real person didn't say, including "cleaned up" versions that change meaning
• An AI image of a real event, person or place presented as a photo
• A document, statistic or study that you can't produce
• A synthetic voice of a real person saying something they didn't say

AI-generated illustrations and dramatizations can be fine when they are clearly labeled and can't be mistaken for reality. The test from the SPJ code is the right one, for journalists and everyone else: could a reasonable person confuse it with the real thing?

Disclosure that means something

Disclosure is for the audience, so it should tell them something. A vague line ("AI may have been used") is worse than useless. It says nothing about what to trust, and it makes the whole piece feel suspect.

Useful disclosure says what the tool did, what it didn't do, and what a human checked. Newsroom examples show the pattern, and the same pattern fits an agency report or an advocacy group's fact sheet.

Placement follows the same logic in newsrooms that do it well: at the end of the piece for research help, directly under the image for illustrations, and on an About or standards page as a general policy. A note about something the audience would want to know belongs where they will see it.

Outlet policies

News outlets set their own rules, and they vary. Some ban AI-generated text outright. Some allow research help but not drafting. Some require disclosure to the editor, not the audience. The same goes for guest columns and letters to the editor: an outlet's submission page often states its AI rule. Editors hate surprises more than they hate tools.

Poynter publishes a starter kit for newsroom AI ethics policies that lays out the decisions: audience-facing uses carry the most risk, back-end reporting help is often encouraged, partnerships with tech companies need disclosure, and the policy should be revisited regularly.

Source privacy: keep it off the cloud

Text pasted into a cloud AI tool goes to a company's servers. What happens next depends on the plan, the settings and the terms that day. As of October 2026:

Even with training off, the text has left your control. Subpoenas, breaches, staff review for abuse and future policy changes all exist. So the rule is plain.

Redacting names isn't enough. Writing style, dates, job details and the shape of a complaint can point straight to one person in a small town. For sensitive material, local software such as Ollama runs open models on macOS, Windows and Linux. Ollama also offers cloud models, so the local setting is the one that keeps text on the machine. Files carry metadata too, and it travels with every upload; AVPrep has a "Strip all metadata" button and can strip location and camera data from photos in bulk.