'Just label it' sounds like advice. In a surprising number of places, it's now the law. AI disclosure law requirements are spreading fast, and they don't all ask for the same thing, which is exactly the problem.
The common thread is simple: synthetic media that could mislead people needs a label. Then the details splinter. Some laws target political content only. Others cover commercial content, like AI-generated product images or fake reviews with synthetic faces. A few reach further and touch any AI-generated likeness used publicly. What counts as a sufficient label varies too. Watermarks, on-screen text, metadata. Pick your poison.
The compliance headache is real, and startups feel it worst. A two-person team ships an AI avatar feature and wakes up needing a fifty-state disclosure matrix. That's not a side quest. That's the product now. A disclosure that satisfies one jurisdiction's wording might not satisfy another's. And these laws are young enough that enforcement is still taking shape, which means the cautious move is to over-disclose, not to guess. Over-disclosure is cheap. Lawsuits are not.
This is exactly where an AI disclosure law tracker earns its keep, and developers building alert features usually reach for an AI disclosure law database API so their tools pull fresh rule changes instead of hardcoding last year's list.
The penalties for getting disclosure wrong are still being written, literally. Most of these statutes are new enough that the first enforcement actions will define what compliance actually costs. Early movers on compliance get to shape that conversation instead of reacting to it.
A synthetic media disclosure law comparison, side by side, is the fastest way to see the pattern. DeepfakeLaw lines up disclosure laws across 50 states, federal law, and 40+ countries, so you can see who requires what, where, and since when.
If your product generates images, video, or voice and you want to know exactly which disclosure rules hit you, I do plain-language compliance audits at deepfakelaw.fyi.