{"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":"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","position":62,"of":73,"minutes":8,"answer":"Disclosure is now a legal duty in several regimes and a contractual one in most institutions, and useful disclosure names which part was generated and who is accountable, placed where a platform cannot strip it.","keyIdea":"Disclosure is now a legal duty in several regimes and a contractual one in most institutions, and useful disclosure names which part was generated and who is accountable, placed where a platform cannot strip it."},"body":{"format":"markdown","text":"## This is no longer only an ethical question\n\nFor several years \"should I disclose that I used AI\" was a matter of personal judgement. It is now, in a growing number of contexts, a legal or contractual duty with consequences. This lesson is the practical map, and like the rest of this module it is a description rather than legal advice.\n\n## What the law requires\n\n**The EU AI Act, Article 50** sets transparency obligations that begin applying in 2026. In outline: systems interacting with people must make clear that the person is dealing with an AI, unless obvious; providers of generative systems must mark synthetic output in a machine-readable way; deployers of deep fakes must disclose that content is artificially generated or manipulated; and AI-generated text published to inform the public on matters of public interest must be disclosed, with exceptions where a human reviews and takes editorial responsibility.\n\n**China** has gone furthest on labelling. Measures effective from September 2025 require both an **explicit label** visible to users and an **implicit label** embedded in file metadata for AI-generated content, with obligations on platforms to check and to label content they detect as synthetic.\n\n**India** has moved through the IT Rules. The government has pursued labelling requirements for synthetically generated information carried by intermediaries, framed within the existing due-diligence regime rather than a dedicated AI statute.\n\n**Advertising and consumer law** applies everywhere already. Endorsements and testimonials must be genuine; the US Federal Trade Commission's rule effective in 2024 explicitly covers AI-generated reviews. Misleading claims about a product are misleading whoever wrote them, and an untrue statement in marketing copy is not excused by its origin.\n\n## What contracts and institutions require\n\n**Academic rules.** Almost every university now has a policy, and they differ sharply: some permit AI use with declaration, some prohibit it for assessed work, some allow it for specified stages. The common failure is assuming last year's policy, or another department's.\n\n**Publishing.** Major journals and the main medical-editor guidance hold that an AI system cannot be an author, because authorship entails accountability, and require disclosure of AI use in the methods or acknowledgements. Many publishers additionally restrict AI-generated images.\n\n**Employment and client contracts.** Increasingly specify whether AI may be used on deliverables, and sometimes require notice. This clause is now standard enough that you should look for it before assuming.\n\n**Courts.** Several jurisdictions now require certification about the use of generative AI in filings, following the fabricated-citation cases.\n\n**Platforms.** Most large platforms require disclosure of realistic synthetic media, and some apply labels automatically from provenance metadata.\n\n## Disclosure that is actually useful\n\nA label that says \"AI was used\" tells a reader almost nothing. Three things make disclosure informative.\n\n**Say which part.** \"The data analysis is mine; the first draft of the summary was AI-generated and edited by me\" is a real disclosure.\n\n**Say who is accountable.** The reason authorship matters is responsibility. Naming the person who checked it is the substance.\n\n**Put it where it survives.** Metadata is stripped by most platforms on upload. If disclosure matters, it goes in the visible caption or the body text, not only in the file.\n\n## Where it is unclear\n\nSpell-check has used statistical models for years and nobody discloses it. Translation, grammar checking, autocomplete and search all sit on the same continuum. There is no principled line, and rules that try to draw one produce absurdities: a policy requiring disclosure of \"AI assistance\" technically covers the predictive text on a phone keyboard.\n\nThe workable test is about **substitution of judgement**, not about tools: did a system perform a cognitive step the reader assumes you performed? If a reader would feel misled to learn how it was made, disclose. If they would shrug, it is a tool.\n\nAnd where a rule exists, follow the rule rather than the test: the institution has already made the judgement, and disagreeing with it privately is not a defence."},"check":{"question":"Which disclosure carries the most information for a reader?","options":[{"text":"A caption on the finished piece reading simply 'Created with AI' in plain type","correct":false,"tempting":"Meets a labelling requirement without telling the reader what was generated or who stands behind it."},{"text":"Provenance metadata embedded in the file, recording the tool that generated it","correct":false,"tempting":"Machine-readable provenance is valuable, and most platforms strip it on upload, so the reader rarely sees it."},{"text":"A note saying which parts were AI-drafted, what was checked, and who answers for it","correct":true},{"text":"A statement naming the commercial model that was used, and giving its exact version number","correct":false,"tempting":"Helps reproducibility but says nothing about which steps were delegated or who checked the result."}],"why":"A reader needs to know what was delegated, whether it was checked, and whom to hold responsible."},"prev":{"slug":"licences-you-actually-have","title":"Reading the licence before you build on it","url":"https://addaly.com/learn/ai-and-you/licences-you-actually-have"},"next":{"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"},"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 you have to say it was AI', in AI, Safety and What Goes Wrong. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-and-you/disclosure-rules-that-bind-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 you have to say it was AI","courseTitle":"AI, Safety and What Goes Wrong","module":{"index":8,"slug":"law-and-ownership","title":"Ownership, law and the public record"},"answer":null,"keyIdea":"Disclosure is now a legal duty in several regimes and a contractual one in most institutions, and useful disclosure names which part was generated and who is accountable, placed where a platform cannot strip it.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"this-is-no-longer-only-an-ethical-question","runs":[{"text":"This is no longer only an ethical question"}]},{"t":"p","runs":[{"text":"For several years “should I disclose that I used AI” was a matter of personal judgement. It is now, in a growing number of contexts, a legal or contractual duty with consequences. This lesson is the practical map, and like the rest of this module it is a description rather than legal advice."}]},{"t":"h2","id":"what-the-law-requires","runs":[{"text":"What the law requires"}]},{"t":"p","runs":[{"text":"The EU AI Act, Article 50","b":true},{"text":" sets transparency obligations that begin applying in 2026. In outline: systems interacting with people must make clear that the person is dealing with an AI, unless obvious; providers of generative systems must mark synthetic output in a machine-readable way; deployers of deep fakes must disclose that content is artificially generated or manipulated; and AI-generated text published to inform the public on matters of public interest must be disclosed, with exceptions where a human reviews and takes editorial responsibility."}]},{"t":"p","runs":[{"text":"China","b":true},{"text":" has gone furthest on labelling. Measures effective from September 2025 require both an "},{"text":"explicit label","b":true},{"text":" visible to users and an "},{"text":"implicit label","b":true},{"text":" embedded in file metadata for AI-generated content, with obligations on platforms to check and to label content they detect as synthetic."}]},{"t":"p","runs":[{"text":"India","b":true},{"text":" has moved through the IT Rules. The government has pursued labelling requirements for synthetically generated information carried by intermediaries, framed within the existing due-diligence regime rather than a dedicated AI statute."}]},{"t":"p","runs":[{"text":"Advertising and consumer law","b":true},{"text":" applies everywhere already. Endorsements and testimonials must be genuine; the US Federal Trade Commission’s rule effective in 2024 explicitly covers AI-generated reviews. Misleading claims about a product are misleading whoever wrote them, and an untrue statement in marketing copy is not excused by its origin."}]},{"t":"h2","id":"what-contracts-and-institutions-require","runs":[{"text":"What contracts and institutions require"}]},{"t":"p","runs":[{"text":"Academic rules.","b":true},{"text":" Almost every university now has a policy, and they differ sharply: some permit AI use with declaration, some prohibit it for assessed work, some allow it for specified stages. The common failure is assuming last year’s policy, or another department’s."}]},{"t":"p","runs":[{"text":"Publishing.","b":true},{"text":" Major journals and the main medical-editor guidance hold that an AI system cannot be an author, because authorship entails accountability, and require disclosure of AI use in the methods or acknowledgements. Many publishers additionally restrict AI-generated images."}]},{"t":"p","runs":[{"text":"Employment and client contracts.","b":true},{"text":" Increasingly specify whether AI may be used on deliverables, and sometimes require notice. This clause is now standard enough that you should look for it before assuming."}]},{"t":"p","runs":[{"text":"Courts.","b":true},{"text":" Several jurisdictions now require certification about the use of generative AI in filings, following the fabricated-citation cases."}]},{"t":"p","runs":[{"text":"Platforms.","b":true},{"text":" Most large platforms require disclosure of realistic synthetic media, and some apply labels automatically from provenance metadata."}]},{"t":"h2","id":"disclosure-that-is-actually-useful","runs":[{"text":"Disclosure that is actually useful"}]},{"t":"p","runs":[{"text":"A label that says “AI was used” tells a reader almost nothing. Three things make disclosure informative."}]},{"t":"p","runs":[{"text":"Say which part.","b":true},{"text":" “The data analysis is mine; the first draft of the summary was AI-generated and edited by me” is a real disclosure."}]},{"t":"p","runs":[{"text":"Say who is accountable.","b":true},{"text":" The reason authorship matters is responsibility. Naming the person who checked it is the substance."}]},{"t":"p","runs":[{"text":"Put it where it survives.","b":true},{"text":" Metadata is stripped by most platforms on upload. If disclosure matters, it goes in the visible caption or the body text, not only in the file."}]},{"t":"h2","id":"where-it-is-unclear","runs":[{"text":"Where it is unclear"}]},{"t":"p","runs":[{"text":"Spell-check has used statistical models for years and nobody discloses it. Translation, grammar checking, autocomplete and search all sit on the same continuum. There is no principled line, and rules that try to draw one produce absurdities: a policy requiring disclosure of “AI assistance” technically covers the predictive text on a phone keyboard."}]},{"t":"p","runs":[{"text":"The workable test is about "},{"text":"substitution of judgement","b":true},{"text":", not about tools: did a system perform a cognitive step the reader assumes you performed? If a reader would feel misled to learn how it was made, disclose. If they would shrug, it is a tool."}]},{"t":"p","runs":[{"text":"And where a rule exists, follow the rule rather than the test: the institution has already made the judgement, and disagreeing with it privately is not a defence."}]}],"check":{"question":[{"text":"Which disclosure carries the most information for a reader?"}],"options":[{"text":[{"text":"A caption on the finished piece reading simply ‘Created with AI’ in plain type"}],"correct":false,"tempting":[{"text":"Meets a labelling requirement without telling the reader what was generated or who stands behind it."}]},{"text":[{"text":"Provenance metadata embedded in the file, recording the tool that generated it"}],"correct":false,"tempting":[{"text":"Machine-readable provenance is valuable, and most platforms strip it on upload, so the reader rarely sees it."}]},{"text":[{"text":"A note saying which parts were AI-drafted, what was checked, and who answers for it"}],"correct":true,"tempting":null},{"text":[{"text":"A statement naming the commercial model that was used, and giving its exact version number"}],"correct":false,"tempting":[{"text":"Helps reproducibility but says nothing about which steps were delegated or who checked the result."}]}],"order":[1,2,0,3],"why":[{"text":"A reader needs to know what was delegated, whether it was checked, and whom to hold responsible."}]},"prev":{"slug":"licences-you-actually-have","title":"Reading the licence before you build on it","minutes":8},"next":{"slug":"when-it-says-something-false-about-you","title":"When a model says something false about a person","minutes":8}}}