Do you have to disclose AI-generated content, and who requires labels?
Increasingly yes, depending on where you publish and what the content is. The EU AI Act makes disclosure of deepfakes a legal obligation, YouTube and TikTok require labels on realistic AI media, Meta labels on detection or self-disclosure, and the US FTC treats AI-driven impersonation as fraud under existing rules. Plain text is the least regulated.
Cite
Global 100 Forum, "Do you have to disclose AI-generated content, and who requires labels?", https://forum.global100.org/q/do-you-have-to-disclose-ai-generated-content-and-who-requires-labels/, accessed 2026-10-11.- Tomas ReyesStaffEdits the deepfakes and provenance sections ·
The legal baseline: EU AI Act Article 50
Article 50 of the EU AI Act is the only broad, binding labelling law in a major market, applying from 2 August 2026 under Article 113. It stacks four duties: chatbots must make clear the user is dealing with an AI; providers of systems generating synthetic audio, image, video or text must mark outputs in a machine-readable, detectable way where technically feasible; deployers who create deepfakes must disclose it, with a lighter duty for evidently artistic or satirical work; and AI text published to inform the public on matters of public interest must be disclosed unless a person holds editorial responsibility for it.
The Commission's Digital Omnibus text gives systems on the market before 2 August 2026 until 2 December 2026 to meet the marking duty.
Platform rules: YouTube, TikTok and Meta
YouTube requires creators to disclose, via the "AI use" setting in YouTube Studio, content that makes a real person appear to say or do something they did not, alters footage of a real event, or generates a realistic scene that never happened. Unrealistic content and minor edits (filters, AI-written scripts, cloning your own voice) need no disclosure. YouTube may label content itself when it carries C2PA metadata, and that label cannot be removed; persistent non-disclosure risks removal or Partner Program suspension.
TikTok requires labels on all realistic AI-generated images, audio or video, auto-labels content carrying C2PA Content Credentials, and does not let creators remove an auto-label. Some AI content is banned regardless: fake crisis events, fake authoritative sources, and the likeness of under-18s or private adults without permission.
Meta's April 2024 policy post chose labels over removal: Facebook, Instagram and Threads apply an "AI info" label on detecting industry-standard AI indicators or on self-disclosure, from May 2024 for organic content. Political and social issue advertisers have had to disclose digital alteration in certain cases since January 2024.
The United States: no label law, but deception rules apply
No general US law requires AI labels; the tool is enforcement against deception. In February 2024 the Federal Trade Commission finalised its rule on government and business impersonation and proposed extending it to individuals, citing voice cloning and deepfakes. Undisclosed AI is not itself illegal there; using it to impersonate a real person or company is.
C2PA: the label that travels with the file
C2PA publishes an open standard, Content Credentials, recording a file's origin and edits in signed metadata; its steering committee includes Adobe, Google, Meta, Microsoft, OpenAI and TikTok, and it is the obvious route to the marking Article 50 demands. Its limits are in the C2PA thread: the data is optional and stripped by screenshots. Text has no container for it and relies on watermarking instead.
Who requires AI labels, on what, and since when
Who requires it What must be labelled From when EU AI Act, Article 50 Chatbots, machine-readable marks on synthetic output, deepfakes, public-interest AI text 2 August 2026; marking grace to 2 December 2026 for existing systems YouTube Realistic content altering real people, events or places, or inventing scenes In force; C2PA auto-labels cannot be removed TikTok All realistic AI-generated images, audio and video; some banned outright In force; C2PA auto-labels cannot be removed Meta Content with industry AI indicators or self-disclosed; altered political ads May 2024; political ads January 2024 US FTC No label duty; impersonation of government, business and (proposed) individuals is actionable February 2024 C2PA Content Credentials Voluntary provenance metadata from tools and cameras Open standard; read by YouTube and TikTok What that means in practice
- Realistic is the trigger. A fantasy scene needs no label; a real-looking person or event does.
- Decide before upload. If the file carries C2PA data, YouTube and TikTok will label it, permanently.
- Text is the gap. Only the EU rule touches AI text, and only for public-interest publishing.
- Disclosure does not launder deception. A label does not make a fake endorsement lawful.
1 more reply
Most helpful first- Tomas ReyesStaffEdits the deepfakes and provenance sections ·
A workflow point for anyone publishing AI-assisted media regularly: export with provenance intact rather than screenshotting the result, and decide up front whether you want the platform's automatic label or would rather self-disclose in your own words. The two are not equivalent. An auto-label from C2PA data is locked; a self-disclosure can be edited and can carry context such as "voice cloned from my own recording". For work heading to EU audiences after August 2026, keeping the machine-readable mark on the file is also the simplest way to show the Article 50 duty was met.
Write something first.
Give people something to work with: at least 30 words on what happened and what you tried.
That is too long. Keep it under 6,000 characters.
Write the question as the title, 15 to 140 characters, no links.
Pick a category.
Add a name (2 to 40 characters, no links).
That email address does not look right.
That was quick. Read the thread, then try again.
The form expired. Reload the page and post again.
Something went wrong with the form. Reload and try again.
Please complete the check and post again.
Limit reached for now. Try again later.
This thread is closed to new replies.
This thread no longer accepts replies.
Something went wrong with the form. Reload and try again.
Post a reply or question first (name and email), then this browser can vote, edit and accept answers.
You cannot vote on your own post.
Only the author (within 30 days) or the forum team can do that.
That email belongs to a forum team account. Use your sign-in link instead.
New accounts are paused for the moment. Try again later.
That was already posted.