Do You Have to Tell Customers They're Talking to an AI Agent? Disclosure Rules Explained

Quick answer

In California, Illinois, and the EU (under the AI Act), disclosing that a customer is interacting with an AI system is a legal requirement in specific contexts, most commonly commercial transactions and any interaction where a reasonable person could mistake the system for a human. Outside those triggers, disclosure is not universally mandated by law, but it is standard practice for any company that wants to keep customer trust and avoid a deceptive-practices complaint. The safest approach is to disclose by default, low-key and early, rather than waiting to see if a specific law applies to your case.

The short answer

Whether you are legally required to tell a customer they are talking to an AI agent depends on your jurisdiction and what the conversation is for. It is not a single global rule. But the direction of travel across every major regulator is the same: assume disclosure will be required soon if it isn't already, and build it in now rather than retrofitting it under a deadline.

Where disclosure is already a legal requirement

California (SB 1001, the "Bolstering Online Transparency" Act). If you use a bot to communicate with a person in California with the intent to mislead them about the fact they're talking to a machine, in order to incentivize a purchase or influence a vote, you must disclose. The trigger is intent to deceive plus a commercial or political purpose, not just "any bot conversation."

The EU AI Act. Article 50 requires that providers ensure natural persons are informed they are interacting with an AI system, unless it is obvious from the circumstances to a reasonably well-informed person. This applies broadly to conversational AI systems deployed in the EU, regardless of whether the conversation is transactional.

Illinois and a growing list of US states. Several states have passed or proposed bot-disclosure statutes tied to specific sectors, most commonly healthcare and financial services, where a customer's belief about who they're talking to affects the weight they give the advice.

Sector-specific rules. Insurance, healthcare, and financial services regulators in multiple jurisdictions already require disclosure when AI is used to interact with a patient or policyholder, independent of any general AI-disclosure statute.

Where it's not strictly required, but you should do it anyway

Most B2B SaaS support interactions, marketing chat widgets, and internal tools fall outside a hard legal trigger today. That does not mean skip it. Three practical reasons to disclose anyway:

1. Trust economics. Customers who discover after the fact that they were talking to an AI, especially about something that mattered, downgrade trust in the brand more than the interaction itself would justify. The reveal is the damage, not the AI.
2. Regulatory direction is one-way. Every jurisdiction moving on this is expanding disclosure requirements, not narrowing them. Building disclosure into your agent now avoids a compliance sprint later.
3. It's cheap to do well. A one-line disclosure at the top of a conversation costs you almost nothing in conversion when it's designed right, and costs you a lot in trust when you're caught skipping it.

Disclosure versus explaining your answers

This is a different problem from source citation. Telling a customer "I'm an AI assistant" is an identity disclosure. Showing them why you gave a specific answer, with sources or confidence indicators, is a trust-in-content problem covered separately in our guide to trustworthy AI agent UI design. You need both, but they solve different failure modes: one is "who am I talking to," the other is "why should I believe what they just said."

How to design disclosure that doesn't hurt conversion

• Disclose in the first message, not buried in a terms link. "Hi, I'm an AI assistant for [Company]" in the opening turn satisfies most disclosure standards and takes one sentence.
• Make human handoff visible and easy, not just theoretically available. Customers tolerate AI a lot better when they can see the exit door.
• Re-disclose after a long gap or a channel switch. If a conversation continues over days or moves from chat to email, don't assume the original disclosure still counts.
• Match the disclosure to the stakes. A shipping-status bot needs a lighter touch than an agent giving healthcare or financial guidance, where the disclosure needs to be paired with a "this isn't professional advice" caveat.

Where this intersects with escalation design

Disclosure isn't just an opening line: it matters most in exactly the moments a conversation gets tense. If a customer becomes upset and starts pushing back hard, good escalation design, covered in our post on the human-in-the-loop approval bottleneck, should include re-confirming they know they're talking to an AI and offering a human immediately, before the interaction sours further.

FAQ

Is there a single US federal law requiring AI chatbot disclosure? No. There is no single federal disclosure statute as of this writing. Requirements come from a patchwork of state laws (California's SB 1001 is the most cited), sector regulators, and increasingly the EU AI Act if you serve EU customers.

Does disclosure apply to internal tools, like an internal IT helpdesk bot? Most disclosure statutes are aimed at consumer-facing commercial or political communication. Internal tools carry lower legal risk but the same trust logic still applies: employees who feel deceived by a bot trust it less afterward.

What counts as "obvious from the circumstances" under the EU AI Act's exemption? Regulatory guidance is still developing, but the working interpretation is narrow: a clearly labeled bot widget with a robot icon and no attempt at a human name or persona may qualify. A named, human-sounding persona without an explicit statement usually does not.

Do disclosure requirements change what my agent's persona can be? Not directly, but they interact with it. If your agent has a human-sounding name and tone (see our post on AI agent personality design), you need a more explicit, not less explicit, disclosure to offset the impression you're creating.