{"course":{"slug":"ai-and-you","title":"AI, Safety and What Goes Wrong","url":"https://addaly.com/learn/ai-and-you","level":"beginner","track":"start"},"lesson":{"slug":"when-it-says-something-false-about-you","title":"When a model says something false about a person","url":"https://addaly.com/learn/ai-and-you/when-it-says-something-false-about-you","position":63,"of":73,"minutes":8,"answer":"Models fabricate damaging claims most reliably about people whose names sit near wrongdoing in public text without having done it, and while liability for the developer is unsettled.","keyIdea":"Models fabricate damaging claims most reliably about people whose names sit near wrongdoing in public text without having done it, and while liability for the developer is unsettled, a user who repeats the claim has published it under ordinary law."},"body":{"format":"markdown","text":"## A new way to be defamed\n\nDefamation law is old and it assumes a publisher. A model producing a false, damaging statement about a named person, to one user, in a conversation nobody else sees, does not fit that assumption neatly, and courts are now working out how it fits.\n\nAn Australian mayor threatened action in 2023 after a model described him as having been convicted in a bribery case in which he had in fact been the whistleblower. A US radio host sued after a model generated a fabricated legal complaint accusing him of embezzlement; a Georgia court granted the developer summary judgment in 2025, reasoning in part about whether a reasonable reader would treat the output as a statement of fact given the disclaimers and the circumstances. A European privacy group filed a complaint in 2025 on behalf of a Norwegian man about whom a model produced a false account of a serious crime, framing the issue not as defamation but as inaccurate personal data under data protection law.\n\nThat last framing may turn out to matter more than defamation. Data protection regimes require personal data to be accurate and give people a right to rectification, and they do not require proof of publication or of reputational damage.\n\n## Why models generate this specifically\n\nThe mechanism is the one from the third module and it explains why the false statements cluster around certain people.\n\nA name plus a domain (a politician, a lawyer, an academic, a journalist) is exactly the kind of prompt where the model has strong statistical associations and weak specific knowledge. It produces the most plausible completion. For a person with a common name, the details of several people merge. For a person who appears in coverage of a scandal in any role (witness, investigator, whistleblower, defence counsel) the association between the name and the wrongdoing is present in the text, and the role can be lost.\n\nSo the systematically most exposed people are those who are named in public material near something bad without having done it. That is a specific, predictable population, and it includes journalists who covered a crime and lawyers who defended someone.\n\n## Who might be liable\n\nUnsettled, and the candidates are these.\n\n**The developer**, as the one whose system produced the words. Defences argued so far include that a conversational output with disclaimers is not a statement of fact, that there is no publication in the traditional sense, and that no intention or negligence exists in the required form.\n\n**The user who republishes it.** This is the clearest case in most legal systems, and the one that should concern you personally. Take a false, damaging statement from a model, put it in an email or a post, and you have published it. Every ordinary defamation principle applies, and \"an AI wrote it\" addresses neither truth nor responsibility.\n\n**The deployer**, where a business puts the output in front of customers: the Air Canada reasoning from the first module.\n\nIntermediary protections that shield platforms from liability for user content were written for hosting other people's speech, and whether they cover speech a system generated is contested.\n\n## If it happens to you\n\n**Capture it properly.** Full screenshots including the prompt, the complete answer, the date, the model and version. Ask again in a fresh conversation to see whether it reproduces, and capture that too. Do not rely on the conversation still being there.\n\n**Use the data protection route where available.** In the EU, India, the UK and other regimes with rectification rights, a request that inaccurate personal data be corrected is faster and cheaper than litigation, and does not require you to prove damage. Address it to the provider's data protection contact.\n\n**Use the provider's process.** Most have a mechanism for reporting factual errors about individuals, and they do act on the ones that reach them.\n\n**Take advice before publicising it.** Amplifying the false statement in order to complain about it is a known way to make the harm worse.\n\n## If you are the user\n\n**Never repeat a model's factual claim about a named living person without independent verification.** The cost of checking is two minutes and the cost of being wrong is somebody's reputation and possibly your own liability."},"check":{"question":"A journalist finds that a model repeatedly describes her as having been charged in a fraud case she in fact reported on. Which route is likely to be fastest and cheapest?","options":[{"text":"A defamation claim against the model provider, since the statement is false and damaging","correct":false,"tempting":"Available in principle but slow and costly, with unsettled questions about publication and whether output is a statement of fact."},{"text":"A complaint to the platform hosting the chat, invoking the intermediary takedown rules","correct":false,"tempting":"Intermediary rules were built for hosting other people's content, and whether they cover generated output is contested."},{"text":"Publishing an article that documents the false output, to set the record straight","correct":false,"tempting":"Understandable, and it risks spreading the false claim to far more people than the conversation reached."},{"text":"A data protection request to rectify inaccurate personal data, which needs no proof of harm","correct":true}],"why":"Rectification needs only that the data is about you and wrong. It requires no proof of damage or publication."},"prev":{"slug":"disclosure-rules-that-bind-you","title":"When you have to say it was AI","url":"https://addaly.com/learn/ai-and-you/disclosure-rules-that-bind-you"},"next":{"slug":"voice-face-and-likeness","title":"Your face and voice as property","url":"https://addaly.com/learn/ai-and-you/voice-face-and-likeness"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Source and assessment review continues. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'When a model says something false about a person', in AI, Safety and What Goes Wrong. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-and-you/when-it-says-something-false-about-you (Accessed: 21 November 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"When a model says something false about a person","courseTitle":"AI, Safety and What Goes Wrong","module":{"index":8,"slug":"law-and-ownership","title":"Ownership, law and the public record"},"answer":null,"keyIdea":"Models fabricate damaging claims most reliably about people whose names sit near wrongdoing in public text without having done it, and while liability for the developer is unsettled, a user who repeats the claim has published it under ordinary law.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"a-new-way-to-be-defamed","runs":[{"text":"A new way to be defamed"}]},{"t":"p","runs":[{"text":"Defamation law is old and it assumes a publisher. A model producing a false, damaging statement about a named person, to one user, in a conversation nobody else sees, does not fit that assumption neatly, and courts are now working out how it fits."}]},{"t":"p","runs":[{"text":"An Australian mayor threatened action in 2023 after a model described him as having been convicted in a bribery case in which he had in fact been the whistleblower. A US radio host sued after a model generated a fabricated legal complaint accusing him of embezzlement; a Georgia court granted the developer summary judgment in 2025, reasoning in part about whether a reasonable reader would treat the output as a statement of fact given the disclaimers and the circumstances. A European privacy group filed a complaint in 2025 on behalf of a Norwegian man about whom a model produced a false account of a serious crime, framing the issue not as defamation but as inaccurate personal data under data protection law."}]},{"t":"p","runs":[{"text":"That last framing may turn out to matter more than defamation. Data protection regimes require personal data to be accurate and give people a right to rectification, and they do not require proof of publication or of reputational damage."}]},{"t":"h2","id":"why-models-generate-this-specifically","runs":[{"text":"Why models generate this specifically"}]},{"t":"p","runs":[{"text":"The mechanism is the one from the third module and it explains why the false statements cluster around certain people."}]},{"t":"p","runs":[{"text":"A name plus a domain (a politician, a lawyer, an academic, a journalist) is exactly the kind of prompt where the model has strong statistical associations and weak specific knowledge. It produces the most plausible completion. For a person with a common name, the details of several people merge. For a person who appears in coverage of a scandal in any role (witness, investigator, whistleblower, defence counsel) the association between the name and the wrongdoing is present in the text, and the role can be lost."}]},{"t":"p","runs":[{"text":"So the systematically most exposed people are those who are named in public material near something bad without having done it. That is a specific, predictable population, and it includes journalists who covered a crime and lawyers who defended someone."}]},{"t":"h2","id":"who-might-be-liable","runs":[{"text":"Who might be liable"}]},{"t":"p","runs":[{"text":"Unsettled, and the candidates are these."}]},{"t":"p","runs":[{"text":"The developer","b":true},{"text":", as the one whose system produced the words. Defences argued so far include that a conversational output with disclaimers is not a statement of fact, that there is no publication in the traditional sense, and that no intention or negligence exists in the required form."}]},{"t":"p","runs":[{"text":"The user who republishes it.","b":true},{"text":" This is the clearest case in most legal systems, and the one that should concern you personally. Take a false, damaging statement from a model, put it in an email or a post, and you have published it. Every ordinary defamation principle applies, and “an AI wrote it” addresses neither truth nor responsibility."}]},{"t":"p","runs":[{"text":"The deployer","b":true},{"text":", where a business puts the output in front of customers: the Air Canada reasoning from the first module."}]},{"t":"p","runs":[{"text":"Intermediary protections that shield platforms from liability for user content were written for hosting other people’s speech, and whether they cover speech a system generated is contested."}]},{"t":"h2","id":"if-it-happens-to-you","runs":[{"text":"If it happens to you"}]},{"t":"p","runs":[{"text":"Capture it properly.","b":true},{"text":" Full screenshots including the prompt, the complete answer, the date, the model and version. Ask again in a fresh conversation to see whether it reproduces, and capture that too. Do not rely on the conversation still being there."}]},{"t":"p","runs":[{"text":"Use the data protection route where available.","b":true},{"text":" In the EU, India, the UK and other regimes with rectification rights, a request that inaccurate personal data be corrected is faster and cheaper than litigation, and does not require you to prove damage. Address it to the provider’s data protection contact."}]},{"t":"p","runs":[{"text":"Use the provider’s process.","b":true},{"text":" Most have a mechanism for reporting factual errors about individuals, and they do act on the ones that reach them."}]},{"t":"p","runs":[{"text":"Take advice before publicising it.","b":true},{"text":" Amplifying the false statement in order to complain about it is a known way to make the harm worse."}]},{"t":"h2","id":"if-you-are-the-user","runs":[{"text":"If you are the user"}]},{"t":"p","runs":[{"text":"Never repeat a model’s factual claim about a named living person without independent verification.","b":true},{"text":" The cost of checking is two minutes and the cost of being wrong is somebody’s reputation and possibly your own liability."}]}],"check":{"question":[{"text":"A journalist finds that a model repeatedly describes her as having been charged in a fraud case she in fact reported on. Which route is likely to be fastest and cheapest?"}],"options":[{"text":[{"text":"A defamation claim against the model provider, since the statement is false and damaging"}],"correct":false,"tempting":[{"text":"Available in principle but slow and costly, with unsettled questions about publication and whether output is a statement of fact."}]},{"text":[{"text":"A complaint to the platform hosting the chat, invoking the intermediary takedown rules"}],"correct":false,"tempting":[{"text":"Intermediary rules were built for hosting other people’s content, and whether they cover generated output is contested."}]},{"text":[{"text":"Publishing an article that documents the false output, to set the record straight"}],"correct":false,"tempting":[{"text":"Understandable, and it risks spreading the false claim to far more people than the conversation reached."}]},{"text":[{"text":"A data protection request to rectify inaccurate personal data, which needs no proof of harm"}],"correct":true,"tempting":null}],"order":[1,2,3,0],"why":[{"text":"Rectification needs only that the data is about you and wrong. It requires no proof of damage or publication."}]},"prev":{"slug":"disclosure-rules-that-bind-you","title":"When you have to say it was AI","minutes":8},"next":{"slug":"voice-face-and-likeness","title":"Your face and voice as property","minutes":8}}}