AI Rules

Only the human part of it, and you have to say which part that is:

• The Copyright Office registers works containing AI-generated material only where a human contributed enough creative expression.
• Applicants must disclose AI-generated content that is more than trivial, and describe the human authorship.
• In its Zarya of the Dawn decision the Office registered the text and the arrangement, but not the AI-generated images themselves.

Usually yes, and you will not see it. It depends which tool made the file:

• ChatGPT: images carry C2PA Content Credentials plus an invisible SynthID watermark.
• Adobe Firefly: Content Credentials are attached automatically when 100% of the pixels came from Firefly.
• Canva: its terms forbid stripping provenance metadata from AI output, which tells you the tags are there.
• The tags travel in the file, so exports, screenshots, and some platforms can drop them.

Often yes, and it turns on where the person is and what the bot is doing:

• California: unlawful to use a bot to push a sale or sway a vote without saying it is a bot.
• Utah: disclose when someone asks, and upfront in some regulated situations.
• EU: the AI Act requires that people be told when they are interacting with an AI system.

There is no single rule that says every AI ad carries a label. What gets you in trouble is leaving out something a customer would want to know, or passing off synthetic people as real. The FTC treats a misleading impression as deceptive advertising, and the EU AI Act adds its own transparency duties for AI-generated content.

Yes, for some content, since August 2, 2026:

• The AI Act’s transparency rules (Article 50) now apply.
• They put two kinds of duties in place. AI providers must mark AI output so it can be detected. Businesses that publish certain AI content must label it, namely deepfakes and some AI-generated publications.
• The European Commission published a voluntary Code of Practice to help companies comply. Meta has said it is signing it.

No, not as a blanket rule. The UK’s ad regulator says there is "no blanket legal requirement in the UK to disclose the use of AI in ads." Its existing codes still apply regardless of how an ad was made, so an AI ad that misleads is treated like any other misleading ad.

The big three all have rules, and each writes its own definition:

• YouTube: disclose realistic altered or synthetic content. YouTube can add the label itself.
• TikTok: label realistic AI content. It also auto-labels some uploads it detects.
• Meta: labels AI images across Facebook, Instagram, and Threads.

Not for being AI-made on its own. Ads get rejected for what they show or claim, and ad platforms apply extra rules to synthetic content in sensitive categories. Google requires election ads to disclose synthetic content that shows people saying or doing things they did not.

Only if it is, and only as far as you can back up. The FTC expects evidence behind performance claims and has brought a run of cases against companies that overstated what their AI does. "AI-powered" on a page that does not use AI is a deceptive claim.

No. The FTC rule on consumer reviews bans fake reviews and testimonials, including ones that misrepresent a real customer experience. Reviews written by insiders have to disclose the connection.

Two separate questions, and the answers differ:

• Copyright: in the US, protection needs human authorship. The Copyright Office says output from a prompt alone is not registrable, while human-authored parts of a work can be.
• Permission: what you may do with the file is set by the tool’s own terms, whether or not anyone holds a copyright in it.

Treat it as no until you check. The FCC has ruled that AI-generated voices in calls count as "artificial" under the robocall rules, which means the stricter consent requirements apply to those calls.

Permission does not transfer with a spreadsheet. Whoever collected those contacts got consent for their own messages, not yours, and US state privacy laws add their own rules on selling and sharing personal information. Check the source of the list before it touches the tool.

Two things decide it, and you need both:

• Your contract with the AI vendor: does it cover this data, and does it keep it out of training?
• Your privacy notice: were the people it describes told their information could be used this way?

Sometimes, and not always in ways you can see:

• Many tools attach Content Credentials, a signed record of how a file was made.
• Some platforms read that record and label the post for you, without asking.
• California’s AI Transparency Act has been in force since August 2, 2026. It makes large AI providers embed provenance data in what they generate and run a free public detection tool.
• It binds the providers, not you. "Covered provider" means a generative AI system with over 1,000,000 monthly users that is publicly accessible in California.
• More is coming: AB 853 extends the same idea to large online platforms and AI hosting platforms on January 1, 2027, and to capture devices like cameras on January 1, 2028.

Often not. Credentials travel in the file, so a screenshot, a re-export, or a platform that strips metadata can drop them. That cuts both ways: your provenance can vanish from your own work, and a file with no credential is not proof a human made it.