Should an AI agent say it is an AI?

This question comes up on nearly every first call, usually phrased as a worry that disclosure will hurt conversion. It is worth separating the ethical answer from the commercial one, because they happen to point the same way.

Our position

The agent says what it is when asked, every time, and there is no configuration that turns that off. This is not a setting we expose and then recommend leaving on. It is not a setting.

It does not volunteer the fact in the first sentence of every conversation, because that would be strange and most customers do not care. But the moment somebody asks, directly or indirectly, the answer is straight.

Why the commercial argument for hiding it is weak

Three reasons, none of which require you to care about the ethics.

• People increasingly ask, and the ones who ask are the ones who already suspect. Denying it to somebody who has guessed correctly is the worst possible outcome.
• The disclosure is rarely what loses the conversation. A slow or wrong answer loses it. A fast, accurate answer from a disclosed agent outperforms a slow one from a person.
• Screenshots travel. A business caught denying it was running an AI has a much bigger problem than a slightly lower conversion rate, and it is a problem that does not go away.

Regulatory direction is one way

Disclosure requirements for automated systems are tightening in several jurisdictions, and nothing on the horizon suggests they will loosen. Building a process that depends on non-disclosure means building something you will have to unwind, probably at an inconvenient moment.

We are not going to tell you what your specific obligations are, because that depends on where you operate and what you do, and a marketing page is the wrong place to get legal advice. What we will say is that designing for disclosure now costs almost nothing and designing against it creates a liability you have to carry.

What disclosure does not require

It does not require the agent to be unpleasant to talk to, to caveat every sentence, or to constantly remind people of what it is. An agent that says 'I'm an AI assistant, but I can help with that' and then helps is doing the job. An agent that opens every message with a disclaimer is just worse to deal with, and that helps nobody.

It also does not require you to name your vendor. The customer needs to know they are talking to software, not the architecture behind it.

The related rule: no invented proof

Disclosure is part of a broader posture that also covers what the agent claims. It does not invent statistics, it does not cite outcomes it cannot support, and it does not describe capabilities you do not have. If it does not know whether you support something, it says so rather than producing a confident guess.

Those two rules together, say what you are and do not claim what you cannot support, cover most of the reputational risk in deploying a customer-facing agent. Neither is technically difficult. Both require somebody to decide they matter before the first conversation rather than after the first incident.