Tell Them It’s AI

An agent that places calls on your behalf raises a question an agent that only answers your own phone never does: the person it reaches didn’t ask to talk to software. Do they have a right to know? States, federal regulators, and the EU have all been answering that question, and they keep landing in the same place. An AI talking to a person has to say it’s an AI. Clearly, and up front. This is what we’ve been reading on where that rule came from and where it’s going.

California wrote the first rule

The first statute to name the idea was California’s B.O.T. Act (SB 1001), in force since July 1, 2019. It makes it unlawful to use a bot to communicate with someone in California while misleading them about its “artificial identity” to incentivize a sale or influence a vote. The escape hatch is disclosure: say it’s a bot, clearly and conspicuously, and the claim falls away. There’s no standalone private right of action — enforcement runs through California’s consumer-protection machinery, including the Unfair Competition Law. But the principle was set. A person dealing with an automated account is entitled to know that’s what it is.

The B.O.T. Act was narrow on purpose. It went after deception tied to commerce and elections, not every chatbot. But it set the template the rest of the country would borrow: don’t hide the machine, and if there’s any chance of confusion, disclose.

Then the rule grew teeth

The 2024–2026 wave moved from “don’t deceive” to “affirmatively disclose,” and widened who can sue. California’s AB 2905, effective January 1, 2025, amends the state’s automatic dialing-announcing device rule: the natural-voice announcement at the top of the call now has to tell the person if the prerecorded message that follows uses an AI-generated or significantly AI-altered voice. California’s SB 243, effective January 1, 2026, requires operators of “companion” chatbots to tell users clearly when a reasonable person could be misled into thinking they’re talking to a human. SB 243 also carries a private right of action for anyone injured by a violation — the strongest enforcement teeth in this area yet.

It isn’t only California. Maine’s Act to Ensure Transparency in Consumer Transactions Involving Artificial Intelligence, effective September 2025, bars businesses from using an AI chatbot that could mislead a consumer into thinking they’re dealing with a human unless the consumer is clearly told otherwise, enforced through the state’s Unfair Trade Practices Act. The pattern across the wave: disclosure is shifting from a defense you raise to a duty you owe.

Washington got there through robocall law

The federal hook is older than AI. The Telephone Consumer Protection Act has long restricted calls that use an “artificial or prerecorded voice.” In February 2024, prompted by a request from 26 state attorneys general and the fake-Biden primary robocall in New Hampshire, the FCC issued a declaratory ruling that AI-generated and cloned voices are “artificial” under the TCPA. AI voice calls now sit under the same consent, identification, and opt-out rules as any robocall.

The FCC then proposed going further. Its August 2024 rulemaking (FCC 24-84) would define an “AI-generated call” and require a clear disclosure at the start of each one that AI-generated technology is being used, plus consent language that specifically covers AI calls and texts. There’s a carve-out for assistive AI used by people with speech or hearing disabilities. The proposal isn’t law yet, but it’s walking the same direction as the states: announce the AI, up front.

Europe drew the brightest line

The clearest statement of the norm is in the EU AI Act. Article 50 requires that people be informed they’re interacting with an AI system unless that’s obvious to a reasonably well-informed person; that providers mark synthetic audio, image, video, and text as machine-readable AI output; and that deployers disclose AI-generated or manipulated media that constitutes a deepfake. The information has to be given clearly, at the latest at the first interaction. These obligations apply from August 2, 2026.

Next to the U.S. patchwork, Article 50 is the broadest version of the same instinct, applied economy-wide rather than vertical by vertical. It has exceptions too, including where the AI is obvious.

The cautionary tale that started it

The episode that made this concrete arrived before the statutes did. At Google I/O 2018, Google Duplex placed human-sounding calls to book a haircut and a restaurant table, complete with “um”s and “mm-hm”s, and never told the businesses they were talking to software. The backlash was immediate. Researchers at the Oxford Internet Institute’s Digital Ethics Lab argued that people should always be able to know that an AI they’re dealing with is not a person, and that blurring the line sows distrust into ordinary interactions.

Within days Google committed to disclosure, and by the time Duplex rolled out later that year it identified itself up front and announced that the call was being recorded. The norm hardened in public opinion years before it was written into law. That’s part of why the law, when it came, kept landing in the same place.

Why disclosure is the right default anyway

There’s an uncomfortable fact under all of this: disclosure costs money. In a field experiment on more than 6,200 outbound sales calls, Luo, Tong, Fang, and Qu (Marketing Science, 2019) found that undisclosed chatbots sold as well as proficient human agents — and that telling the customer it was a bot before the conversation cut purchase rates by more than 79.7%. People got curt. They judged the bot less knowledgeable and less empathetic. The honest move is the expensive one.

That gap is exactly why disclosure has to live in the floor of the system and not in a growth dashboard. Left to incentives, the disclosure gets quieter every quarter. Written into the product as a default with no per-call toggle, it stays. The law is converging on the same instinct our commit policy starts from: an agent that acts in your name should be accountable, and accountability begins with the other person knowing who, or what, they’re talking to.

What we build to

For an agent that places calls for a business, disclosure is what makes the delegation trustworthy instead of a trick. Our agent says it’s AI at the start of every call and again if anyone asks, and that floor isn’t a setting. The law is still being written, the circuit courts are still arguing about consent, and the EU clock starts in August. But the smoothest-talking voice in the room is exactly the one that should tell you what it is. California thinks so. The FCC thinks so. The EU thinks so. So do we: tell them it’s AI.