This one comes up in almost every AI agent project we scope: a chat widget goes on the website, or an AI phone agent starts picking up the line, and somewhere in the build someone asks, “wait, do we have to tell people this is AI?” It’s a fair question, and the honest answer has two parts. Legally, it depends on where your customers are and what the AI is doing. Practically, you should just disclose it well regardless, because it’s cheap insurance against the one outcome that actually hurts you: a customer feeling tricked. This post covers both parts. It’s general information based on current law as of September 2026, not legal advice for your specific business — talk to a lawyer if you want that.

The EU AI Act is the clearest rule that exists right now

If any of your customers are in the EU, this is the one to know. The EU AI Act’s transparency rules, in Article 50, require that people be told when they’re interacting with an AI system — a chatbot, a voice agent, an AI avatar — and that this notice come before or at the very start of the interaction, in a way that’s clear and easy to notice, not buried in a terms page. It doesn’t matter whether the underlying AI is “high-risk” under the rest of the Act; this obligation applies regardless.

Article 50 also covers AI-generated content: audio, images, video, or text that’s synthetically generated or manipulated has to be marked in a way that’s machine-readable and technically detectable as AI-made.

These obligations become enforceable on August 2, 2026, with a four-month transition (to December 2, 2026) for systems that were already on the market before that date, giving businesses a little runway to adjust. Penalties can run up to €15 million or 3% of worldwide annual turnover, whichever is higher — and the law doesn’t require you to have an EU office to be in scope. If your chatbot or AI-generated content reaches people in the EU, it can apply to you.

What the US actually requires — it’s patchier, and mostly state-level

There’s no single US federal law that says “label your chatbot.” What exists instead is a growing, uneven set of state rules, plus a general federal backstop.

California has had a bot disclosure law on the books since 2019 (the B.O.T. Act, SB 1001). It makes it unlawful to use a bot to communicate with a California resident in order to incentivize a sale or influence how they vote, without disclosing that it’s a bot. The disclosure has to be “clear, conspicuous, and reasonably designed to inform” — there’s no fixed script, but it can’t be hidden in fine print. There’s no private right of action; it’s enforced by the state Attorney General under California’s unfair competition law.

Utah’s AI Policy Act takes a different approach: the baseline rule is that a business using generative AI to communicate with a consumer has to disclose that fact if the consumer directly asks. A 2025 amendment added a stricter, proactive requirement for higher-stakes interactions — ones involving health, legal, or financial advice — where disclosure has to happen at the start of the conversation, not just on request. Utah also offers a safe harbor: businesses that clearly disclose AI involvement upfront and throughout get some legal protection under the Act.

A handful of other states — Colorado, Connecticut, Maine among them — have been adding their own disclosure requirements, particularly for companion-style or customer-service chatbots, and more state bills are moving through committees. This is a moving target; if you operate in multiple states, it’s worth a periodic check rather than a one-time read.

Underneath all of that sits the FTC. There’s no dedicated federal chatbot-disclosure statute, but Section 5 of the FTC Act bars unfair or deceptive practices, and the FTC has been explicit that a chatbot designed to pass itself off as a human, in a way that misleads a consumer, can qualify. That’s a lower, general-purpose bar rather than a specific rule with a specific label requirement — but it means “our AI never technically said it was human” isn’t necessarily a safe position if the overall effect on a reasonable customer is deceptive.

Legal requirement aside, why disclosure is just good practice

Set the specific laws down for a second, because the more useful question for most small businesses is simpler: does telling someone they’re talking to AI help or hurt the interaction? Mostly, it helps, for a few concrete reasons.

• It sets expectations correctly. A customer who knows they’re talking to AI adjusts how they phrase things and doesn’t get frustrated when the system can’t make a judgment call a human would.
• It makes the eventual reveal a non-event. Customers figure out they were talking to a bot eventually, either mid-conversation or after. If you told them upfront, that’s a shrug. If you didn’t, it reads as a small deception, even if you never explicitly claimed to be human.
• It’s a trust signal, not a trust cost. Being upfront about using AI, paired with the AI actually being good at its job, tends to read as competent and modern rather than cheap. The trust damage comes from AI performing badly or lying about what it is, not from AI existing.
• It gives people an obvious exit to a human. Disclosure works best paired with an easy way to ask for a person. That combination — “you’re talking to AI, and here’s how to reach a human” — covers most of what regulators are actually trying to accomplish.

Practical ways to disclose, without making it a whole production

None of this requires a legal disclaimer page. A few low-effort patterns cover almost every case:

• Chat widgets. Label it in the interface itself — “AI Assistant” or “AI Agent” in the widget header, plus a first message like “Hi, I’m an AI assistant. I can help with X and Y, and I can connect you with a person any time.” That single opening line satisfies the “before or at the start” timing that the EU AI Act asks for and covers the “on request” bar most US state laws set.
• AI phone agents. A short line at the start of the call — “You’ve reached [business], I’m an AI assistant” — before getting into the reason for the call. It costs about two seconds of call time and removes any ambiguity, which matters more on voice than on chat, since tone and pacing on the phone can pass as human much more easily than typed text.
• AI-generated content on your site. Blog posts, images, or marketing copy that were substantially AI-generated are increasingly expected to carry some indication of that, particularly under the EU AI Act’s content-labeling rules. A simple footer note or byline convention handles this without turning every page into a disclaimer.
• Keep the escape hatch visible. Whatever the channel, make “talk to a person” a standing option, not something a customer has to fight for. Regulators and customers care about this almost as much as the initial label.

None of that requires legal review to implement. It’s mostly a copywriting and UI decision, made once, at the point you build or configure the AI agent.

Where it gets genuinely higher-stakes

Disclosure matters more in some contexts than others, and it’s worth being more careful in a few specific ones: anything touching health, legal, or financial advice (exactly the areas Utah singled out for stricter, proactive disclosure); anything where the AI is making a decision that affects someone materially, like pricing, eligibility, or an approval; and anything where a reasonable person might act differently if they knew a human wasn’t involved, like a negotiation or a complaint escalation. If your AI agent sits anywhere in that territory, treat disclosure as non-negotiable rather than a nice-to-have, and consider getting a lawyer’s eyes on the specific wording.

Where this leaves you

The regulatory picture will keep shifting — the EU AI Act’s transparency rules take full effect this year, and US states are still actively adding to the pile. But the practical answer for most small businesses doesn’t really depend on tracking every one of those changes: label your chatbot, greet callers honestly on an AI phone line, and keep a human reachable. That single habit satisfies the strictest version of every law discussed here, costs you almost nothing to implement, and is simply the version of AI customer service that holds up over time.

If you’re building or configuring an AI agent for your business and want the disclosure baked in correctly from the start rather than bolted on later, that’s part of what our AI agent development work covers. Tell us what you’re building and we’ll help you get the wording right for your actual setup.

Frequently Asked Questions