The Customer Who Tells You They're a Competitor, or a Journalist

Quick answer

When a user explicitly states they work for a competitor or are a journalist researching a story, do not treat the disclosure itself as adversarial intent. The right default is to answer honestly within the same boundaries the agent already applies to everyone, decline to share anything genuinely confidential, and avoid two failure modes at once: being cagier or more defensive than the agent would be with an ordinary customer, which reads as having something to hide, and volunteering competitively or reputationally sensitive information the agent would never share with a customer just because the person was polite enough to disclose who they are.

Disclosed identity and detected intent are different problems

Detecting an undisclosed adversarial user, someone probing for competitive intelligence or looking for a story while presenting as an ordinary customer, is a pattern-recognition problem covered in adversarial audience detection and response policy. That post is about spotting behavioral and conversational signals that suggest a user's stated context does not match their actual intent.

A user who states upfront, "I'm a reporter looking into how your AI agent handles refund disputes" or "I work for a competing company and I'm curious how you handle X," has removed the detection problem entirely. They are telling you exactly who they are and, often, exactly why they are asking. The open question is not whether they are adversarial, it is what the agent should do differently, if anything, once that context is known.

Default to the same honesty and boundaries as any other user

The instinct to become more guarded with a self-disclosed journalist or competitor is understandable but usually counterproductive. A visibly defensive or evasive response to a straightforward question, especially one a normal customer could ask and get answered plainly, signals that there is something worth hiding, which is often a worse outcome than just answering the question honestly within existing limits. The agent's baseline policy boundaries, covered in designing a refusal boundary for business rules, should already define what is and is not shareable with any user. A self-disclosed journalist or competitor employee does not change those boundaries; it just makes the audience for the answer explicit.

Concretely: if a customer asking the same question would get a straightforward answer, a self-disclosed journalist or competitor employee asking it should get the same straightforward answer. If a customer asking the same question would get a boundary-respecting decline (because the answer involves confidential pricing logic, internal architecture, or another customer's data), the journalist or competitor should get that same decline, delivered the same way, not a more elaborate or suspicious-sounding version of it.

Where disclosure should change behavior: internal routing, not the answer itself

The one place self-disclosure should reasonably change agent behavior is routing and internal awareness, not the substance of the response. A conversation where a user has identified as a journalist working on a story is worth flagging for a human on the comms or PR side to be aware of, similar in spirit to the escalation logic in detecting and de-escalating an upset customer mid-conversation, where certain conversation types warrant human visibility even when the agent itself is handling the immediate response competently. The agent does not need to change what it says; the company benefits from knowing the conversation happened.

Do not build a policy that rewards non-disclosure

A subtle risk in over-restricting responses to self-disclosed adversarial users is that it creates an incentive to not disclose. If a competitor or journalist gets a noticeably more useful answer by simply not mentioning who they are, the policy is training exactly the wrong behavior, and it undermines the honesty covered in truth in marketing for AI agent capability claims: a company that treats honest disclosure as a reason to be less forthcoming is quietly signaling that its normal answers are more generous than they should be, since it only tightens up once it knows who's watching.

FAQ

Should the agent verify a self-disclosed journalist or competitor claim before responding differently? Generally no, since the recommended default is to respond the same way regardless of the claim's authenticity, using existing policy boundaries. Verification effort is better spent on internal awareness and routing than on gatekeeping the response itself.

Is it ever appropriate to decline to answer specifically because someone identified as a journalist? Only if the underlying information itself would be declined for any user, not because of who is asking. Declining specifically because of the journalist disclosure, for information that would otherwise be shared, is the pattern to avoid.

How does this interact with undisclosed adversarial audience detection? They are complementary, not overlapping. Detection policy handles users who conceal adversarial intent; this policy handles users who state it openly, and the two should not converge on the same suspicious-by-default treatment.

Should self-disclosure trigger any internal notification? Yes, this is the one place disclosure should change behavior. Flagging the conversation for human visibility, particularly for journalist inquiries, is reasonable even when the agent's own response does not change.