AI audio has to say it's AI now. Here's the law.

Disclosure is becoming a rule, not a courtesy

For a few years, telling your audience that a voice was AI-generated was a matter of taste. Some creators did it, most did not, and nobody made you. That window is closing. Across 2025 and 2026, disclosure has been hardening from etiquette into law on both sides of the Atlantic, and the deadlines are now close enough to plan around.

This is not a scary post about a crackdown. It is a map, because the rules are a patchwork and a patchwork is easy to get wrong. If you make audio with any AI voice tool, ours or anyone else's, this is the terrain you are standing on.

Europe set a date: 2 August 2026

The clearest line in the sand is the European Union's AI Act. Its Article 50 puts transparency duties on the people who make and deploy synthetic media, and those duties apply from 2 August 2026. Two things matter for audio. First, AI-generated audio is meant to carry machine-readable marking, a signal in the file itself that says this was synthesized. Second, when audio is a deepfake, a clone or imitation of a real, identifiable person, whoever publishes it must disclose that plainly to listeners.

It binds both the people who offer the tool and the people who publish the output, and using a compliant engine does not transfer the duty away. The Act is European, but the internet is not, so anyone shipping audio to EU listeners is inside its reach.

America's answer is a patchwork

The United States has no single federal disclosure law for synthetic audio, but it is far from a free-for-all. More than two dozen states have passed deepfake laws, most aimed at elections, that either restrict synthetic political audio near an election or require a conspicuous label saying the audio was artificially generated.

Two federal moves sharpen the edges. The FCC ruled in 2024 that AI-generated, cloned voices in phone calls fall under the existing robocall rules, so a synthetic voice on an unconsented call is already illegal. And the FTC's impersonation rule, built to fight scams that mimic businesses and officials, is being extended toward the impersonation of individuals. The throughline holds even where the statutes do not match: synthetic audio that could deceive someone is where the law gets strict. Our deeper read of these rules is in deepfake disclosure laws.

What this means if you make AI audio

Most honest creative uses, a narrated article, a video voice-over, a meditation track, are not what these laws are hunting. The risk concentrates in a few clear places, and they are worth knowing by heart.

• Anything that imitates a real, identifiable person is the high-risk zone. That is a deepfake under the EU rules and the heart of the US right-of-publicity and impersonation laws.
• Political and election audio carries its own disclosure rules in a growing list of states; assume a label is required.
• Synthetic voices in phone calls are robocall-regulated; consent and identification are not optional.
• Everywhere else, a simple note that the audio is AI-generated is rarely required yet, but it is cheap insurance, increasingly expected, and the direction every one of these laws points.

Where Cantari stands

We did not wait for a deadline to take a side. Audio shared to our community already carries an AI-generated label, our terms ask you to keep that disclosure in place when you publish, and we treat marking generated output as a duty we hold, not one we push entirely onto you. The direction of travel, machine-readable provenance riding along with the file, is one we intend to meet rather than dodge.

There is a reason this fits us. The whole blog is about not faking things: a benchmark we cannot tilt, samples we verify twice, audio that is yours to keep. Saying plainly that synthetic audio is synthetic is the same instinct, pointed at your audience instead of ours.

A label is not an apology

It is tempting to read a disclosure rule as a verdict, as if the law is calling AI audio second-rate and making you admit it. That is the wrong frame. A label is not an apology for the work; it is a courtesy to the listener and a shield for the maker. The creators who get burned by the coming rules will be the ones who hid the ball, not the ones who said plainly what they made.

So treat disclosure the way we treat our own numbers: not as a confession, but as the thing that lets people trust you. Make great synthetic audio, and say that it is synthetic. Those two facts have never been in tension, and the law is finally catching up to that.