AI-disclosure law

01One practice, applied everywhere

Dohos's responsible-AI policy requires a clear disclosure, at the start of an interaction and again after any material reconnect, that the caller is talking with an AI system — stated in understandable language, alongside accurate identification of which restaurant the call is on behalf of and a real human or restaurant fallback. That requirement does not vary by state, city, or how any specific jurisdiction has or hasn't legislated AI disclosure yet. The exact disclosure requirements are set out in full at the AI and Voice Transparency Notice; this page doesn't restate them.

02Why a single practice rather than a jurisdiction toggle

AI-disclosure law is a genuinely unsettled area — new state and local statutes are still being introduced and taking effect, each with its own specific trigger, wording requirement, and scope. A per-jurisdiction toggle would need to be correctly updated every time a new one passed, and would fail exactly where it matters most: a caller in a not-yet-covered jurisdiction getting no disclosure at all because the toggle hadn't caught up yet.

Applying the same disclosure everywhere sidesteps that failure mode by construction. It also means the practice does not depend on correctly identifying a caller's exact location before deciding whether to disclose — the same fail-closed instinct that governs Dohos's approach to jurisdiction generally, applied here to disclosure specifically rather than to whether a jurisdiction is served at all.

03What the disclosure does not claim

Dohos's AI governance policy is specific about language to avoid in describing what the system does: no claim of "error-free," "unbiased," "fully automated," "human-level," "compliant," or "never trained on [some data]" without precise, current evidence behind the exact words used. Applied to this page: the disclosure practice being applied uniformly and ahead of a specific jurisdiction's own deadline is not the same claim as "compliant with every AI-disclosure law," and this page does not make the second claim.

• the disclosure identifies that the interaction is automated — it does not claim the system understands a request the way a person would
• a natural-sounding voice is not treated as evidence that the disclosure requirement has been satisfied by itself — the actual words disclosed are what matters
• the practice does not vary based on an unverified guess at the caller's jurisdiction

04No hidden AI, no impersonation

The disclosure obligation runs in both directions. Dohos's AI governance policy prohibits hidden AI identity or impersonating a restaurant employee or any other person outright — not merely discouraging it. The system does not claim to be a restaurant employee, owner, or manager, and if a human does join the call, the caller is told whether that person represents the restaurant or Dohos where the distinction matters.

That prohibition sits underneath why the disclosure is credible rather than a formality: the system's authority is genuinely bounded before it ever gets to the disclosure. Models may propose a response or an action, but deterministic checks — not the model itself — validate identity, authority, the exact menu and price version, and payment state before anything material happens. A caller who is told they're talking to an automated system is being told something true about how little independent authority that system actually has, not just informed that a machine is on the line.

The disclosure also identifies which restaurant and location the call is on behalf of, accurately, at the same point it identifies the system as automated — a caller should never be uncertain whether they've reached the right place because a generic AI greeting stood in for a specific one.

05Keeping up with new law

New AI-disclosure statutes are reviewed against Dohos's standing practice as they take effect, the same way every other policy behind this compliance cluster reviews itself on a material-change trigger rather than a fixed calendar. Where a specific statute would require something the standing practice doesn't already do, that is a product and legal change to make, not a fact to announce here ahead of making it.

06Where the adjacent obligations live

AI-disclosure law is one of several call-opening obligations Dohos treats as related but distinct: state recording law governs whether the call may be captured, the TCPA governs whether and how the call may be placed at all, and accessibility conformance governs whether the disclosure itself is usable by the person receiving it. None of the four is treated as satisfied by satisfying another.

Related documents

• AI and Voice Transparency Notice — the full disclosure content this page's practice describes at a design-reasoning level.
• State recording law — the adjacent, distinct obligation governing whether a call may be captured.
• TCPA — the adjacent, distinct obligation governing whether a call may be placed.
• Accessibility conformance — whether the disclosure itself is usable by the person it's delivered to.