Do you have to tell people they are talking to an AI? What the rules say

Published by 3Gen Internet Corporation, maker of OpenAssistant, Skej and AGNT. Comparisons use public documentation, without hands-on testing.

Two ceramic conversation bubbles linked by a clear identity badge and a small legal balance scale, an open rulebook and a plain notification card beside them, transparent AI communication.

This is not legal advice. We explain what the linked primary sources (statutes, regulator pages and the EU’s official text) say, including their stated legal status. Rules can change; confirm their current status with the official sources. We do not say that any rule applies to your business. That depends on where you operate, who you contact, how large you are and what your AI does. The last section is a list of questions to take to a lawyer.

The short answer

There is no single rule that covers every AI in every setting. The picture is a patchwork:

US federal. The FCC treats AI-generated voices as “artificial” voices under the phone-consent law, so consent comes first. That is a consent rule, not a rule to announce yourself. The sources we read show no final FCC rule requiring AI call disclosure, and no FTC page setting a general rule that a customer-facing agent must say it is AI.
States. California and Utah have disclosure rules with different triggers (a sale, a direct question, a misled customer). Colorado’s re-enacted act brings notice duties from 2027-01-01.
EU. Article 50(1) of the AI Act applies from 2 August 2026, and the Omnibus changes did not delay it.

Because the triggers differ, a default of saying so up front is easier to run than a map of exceptions. Whether a given rule requires it for you is a question for counsel. Some products build the habit in: OpenAssistant’s page says it “never pretends to be you unless you tell it to” (OpenAssistant). That is the vendor’s statement, not a legal conclusion.

The rules, one by one

Confirm the current status with the linked official sources before relying on a rule.

Rule	Who it covers	What it requires	Status	Source
FCC Declaratory Ruling 24-17 (phone consent law)	Callers using AI-generated voices	Counts such calls as “artificial or prerecorded voice”. Needs the called party’s prior express consent, and prior express written consent for advertising or telemarketing. Not an “announce you are AI” rule.	In force. FCC 26-16 (a March 2026 proposed rulemaking on call centers) says it does not intend to revisit the point. FCC 26-67 (adopted 2026-09-30) changes consent revocation rules; its comment dates are tied to Federal Register publication, which we did not see.	FCC 24-17, FCC 26-16, FCC 26-67
47 CFR 64.1200(b)(1), as quoted in FCC 24-17	Artificial or prerecorded voice messages	Must state clearly the identity of the business or person responsible. That is the caller’s identity, not AI status.	Quoted in the 2024 ruling; we did not open the CFR text.	FCC 24-17
AI call disclosure	Call centers	FCC 26-16 describes a Senate bill (S.2495) that would require a call center to disclose when a consumer is talking to AI. The sources we read show no final rule.	Proposed bill only.	FCC 26-16
California SB 1001 (Bus. & Prof. Code 17941)	Use of a bot online with a person in California, with intent to mislead them about its artificial identity to incentivize a sale or influence a vote	Not liable if the person using the bot discloses that it is a bot. Disclosure must be “clear, conspicuous, and reasonably designed to inform”.	In force since July 1, 2019.	Statute
California SB 243	Companion chatbots	A “not human” notice if a reasonable person would be misled to believe they are talking to a human. The definition excludes a bot “used only for customer service”, among other uses.	Chaptered Oct 13, 2025. Operative date not confirmed; operator reporting starts July 1, 2027.	Bill
California AB 1609	“Large private business”: over $500 million gross annual revenue nationally; “customer” means a natural person residing in California	May not represent a customer service chatbot as human. Clear disclosure where a reasonable person would likely be misled. A feature to request a human agent, with a good-faith effort to connect within 15 minutes or book an appointment within one business day. Public prosecutors enforce; civil penalty up to $5,000 for an initial violation and $10,000 for each later one; no private right of action.	Approved Sept 28, 2026 (Chapter 733). Operative date not confirmed. The text mentions channels not offered “as of January 1, 2027”, which suggests a 2027 start.	Bill
California AB 316 (Civil Code 1714.46)	Defendants in cases about harm caused by AI	May not assert that the AI “autonomously caused the harm”. Other defenses stay available. Not a disclosure rule, but it answers “the agent did it on its own”.	Approved Oct 13, 2025.	Bill
Utah Code 13-77-103 and 13-77-104	A supplier using generative AI with a consumer	Disclose that it is AI if the person makes “a clear and unambiguous request”. Regulated occupations disclose high-risk interactions up front. No enforcement action if the AI says so up front and throughout.	Version effective 2025-05-07. The Utah site labels that page “Historical Code”, so we could not confirm it is the current text or whether a sunset or 2026 amendment applies.	13-77-103, 13-77-104
Colorado SB 26-189 (automated decision-making technology)	Developers and deployers; consequential decisions including employment	Deployers must give consumers clear and conspicuous notice at the point of interaction with a covered tool; developers owe deployers documentation. Attorney general enforces; 60 day cure notice before Jan 1, 2030.	Signed 2026-05-14, repealing and re-enacting SB 24-205. The final act text (the 05/12 version, PDF) says the act takes effect January 1, 2027. We read that version and the bill summary, not the signed copy.	Bill page
EU AI Act Article 50(1) and 50(5)	Providers of systems meant to interact directly with people	Design the system so people are informed they are interacting with an AI, unless obvious. At the latest at the first interaction.	Applies from 2 August 2026. Regulation (EU) 2026/1744 (the Omnibus) did not delay it.	Regulation 2024/1689, Omnibus
EU AI Act Article 50(2) and 50(4)	Providers of generative systems (marking); deployers (deep fakes)	Machine-readable marking of AI-generated content; deep fake disclosure.	Systems already on the market before 2 August 2026 have until 2 December 2026 for 50(2) marking.	Omnibus

The FTC is a different case. Its pages set no general disclosure rule that we read. Its 2024 “Operation AI Comply” announcement quoted the then-Chair saying “there is no AI exemption from the laws on the books” (FTC). It set aside its Rytr order on 2025-12-22 while saying it would still pursue those who use AI to deceive (FTC). On 2026-03-24 it announced a proposed settlement with Air AI over alleged false claims about earnings and refund guarantees (FTC). On 2026-07-01 it sought comment, until July 31, 2026, on a proposed policy statement about AI companies that distort their systems’ outputs for undisclosed ideological aims. It is not about telling people they are talking to an AI, and we saw no final version (FTC). CAN-SPAM requires commercial email to say it is an ad, which is a different thing from saying an AI wrote it (FTC guide).

What the table hides

The triggers differ. California SB 1001 turns on misleading someone to make a sale. Utah turns on a direct question. AB 1609 turns on a reasonable person being misled. The EU turns on the first interaction.
Providers and deployers. EU Article 50(1) is written as a provider duty. If you buy an agent, ask the vendor how it is built to inform people. That does not tell you what you owe as the business using it.
Hiring is separate. Laws on AI in recruiting are in our hiring law guide.

What vendors say about disclosure

Little, in the pages we read. HeyMilo says candidates “are always told” they are talking to an AI interviewer (HeyMilo). 11x says the lawful basis for contacting a person on a given channel is the customer’s (11x docs), and says its Expansion product calls customers who opted in to AI calls (11x). Artisan says Ava “cannot legally make calls” (Artisan) and sends in your reps’ names from your own sending domains (Artisan pricing). Regie.ai’s terms say you are solely responsible for calling-law compliance (Regie.ai terms). For the other vendors we reviewed, we did not record any statement about AI disclosure.

A disclosure policy template

Adapt this with counsel. It is a starting point, not a legal standard.

Where to say it

Chat. In the first message, plus a visible label on the widget.
Email. In the signature and the opening line of a first message. If a tool sends under a person’s name, decide in writing whether the message says an AI was involved.
Voice. In the first sentence, before any questions, along with the identity of your business. Get consent before AI-voice calls begin.
Text. In the first message of a thread.
Any time someone asks. Answer plainly in one sentence. Never deflect.
Every channel. Offer a way to reach a person, and record how fast you respond.

Wording options

Chat: “You’re chatting with an AI assistant for [Company]. Ask for a person at any time.”
Voice: “Hi, this is an automated AI assistant calling on behalf of [Company].”
Email footer: “An AI assistant working for [Company] wrote this message. Reply and a person will pick it up.”
If asked: “Yes, I’m an AI. I can bring in a person if you’d like.”

Rules for your team

Give the policy one owner and a review date.
Do not give an AI a human name, photo or backstory meant to make people think it is a person.
Log the exact wording used, by channel, and when it changed.
Test it: send a message as an outsider and check the disclosure appears.

Questions for a lawyer

Which of the rules above could reach us, given where our customers, prospects and staff are?
Are we a “provider” or a “deployer” under the EU Act, and what does Article 50 mean for us?
Does AB 1609’s revenue threshold apply to us, and when is it operative?
If an AI sends email in a rep’s name, does the message need to say so?
Do our AI voice calls have the consent they need, and how do we handle revocations after FCC 26-67?
What do state call-recording and telemarketing laws add? We did not research them.
Does our vendor contract say who is responsible for disclosure wording?
How should we word and store our records of disclosure?

Frequently asked questions

Is there one rule that says you must tell people it is an AI?

No. There is no single rule that covers every AI in every setting. The picture is a patchwork: FCC consent rules for AI-generated voices, California and Utah disclosure rules with different triggers, Colorado notice duties from 2027-01-01, and Article 50(1) of the EU AI Act from 2 August 2026.

Does the FCC require AI calls to announce themselves?

The FCC treats AI-generated voices as artificial voices under the phone-consent law, so consent comes first. That is a consent rule, not a rule to announce yourself. The sources we read show no final FCC rule requiring AI call disclosure.

Is it safer to say so up front?

Because the triggers differ, a default of saying so up front is easier to run than a map of exceptions. Whether a given rule requires it for you is a question for counsel.

Do AI vendors disclose that the AI is an AI?

Little, in the pages we read. HeyMilo says candidates are always told they are talking to an AI interviewer. This guide is not legal advice.

Try OpenAssistant

If you want an assistant whose own page says it never pretends to be you unless you tell it to, Try OpenAssistant. It has a free Starter plan, and the paid plans are on the OpenAssistant pricing page.