Do AI Calls Have To Say They're AI? Disclosure Rules in 2026

The FCC settled the federal question in 2024

In February 2024, the FCC issued a declaratory ruling that voices generated by AI count as "artificial or prerecorded voice" under the Telephone Consumer Protection Act. That wasn't a new law — it was the FCC saying the existing one already covers AI. The consequences are immediate for anyone running voice agents:

• Outbound AI-voice calls need the prior express consent the TCPA has always required for artificial-voice calls — and for telemarketing, that means prior express written consent.
• The TCPA's identification rules apply: the call must identify who is calling and provide contact information.
• The private right of action applies: $500 per violating call, $1,500 if willful.

So the federal layer isn't a disclosure mandate per se — it's a consent regime that treats an AI voice like a robocall. The explicit "tell them it's AI" requirements are coming from the states.

States are writing "say you're AI" into law

State legislatures have moved fast since 2024, along two tracks. Some states amended existing robocall statutes: California's AB 2905 (effective January 1, 2025) requires robocalls using an AI-generated or significantly AI-altered voice to disclose that fact at the start of the call. Others passed AI-specific consumer-protection laws that reach voice agents in commerce: Maine's AI transparency law (2025) bars using an AI chatbot in trade in a way that could mislead a consumer into believing they're dealing with a human, and Utah's AI Policy Act requires disclosure in generative-AI consumer interactions — with a safe harbor when the AI clearly identifies itself as AI from the outset.

The pipeline behind them only points one direction: Colorado's AI Act takes effect in 2027, more state bills are in progress, and the FCC has an open proceeding on requiring in-call AI disclosure at the start of every AI-voice call. The patchwork changes every legislative session — and each change adds disclosure, never removes it.

Tracking which states allow silence is a losing game. The list you verify today is stale after the next session, the called party's state is what controls, and — as with recording consent — you often can't know that state reliably from the number. The durable posture is the same one that resolves every patchwork problem in calling compliance: satisfy the strictest rule everywhere, and disclose on every call.

What a good disclosure sounds like

A compliant, trust-preserving disclosure is short, early, and unambiguous:

• Early: in the opening line, before the substance of the call. A disclosure the caller only reaches if the call goes long isn't a disclosure.
• Plain: "AI assistant" or "virtual assistant" — words a normal person parses instantly. Not "automated digital experience."
• Attributed: whose behalf the call is on, which also satisfies the TCPA's identification requirement in the same breath.

The conversion-rate objection, addressed

The standing worry is that admitting the caller is AI kills the conversation. The experience of teams running disclosed agents at scale points the other way: people increasingly recognize synthetic voices within a few sentences, and the damaging moment isn't the disclosure — it's the caller realizing the voice was pretending. A disclosed agent that is fast, useful, and polite keeps exactly the conversations that were ever going to convert. The ones disclosure loses are the ones deception would have lost later, angrier, and with a screenshot.

Why disclosure should be a default, not a feature

Where a compliance rule lives determines whether it actually happens. A disclosure that each customer must remember to paste into each agent's prompt will be missing from some meaningful fraction of agents forever. A platform default inverts that: every new agent discloses unless someone deliberately turns it off.

That's how Vosy ships it: AI disclosure is a per-agent setting that defaults to ON, with a configurable disclosure line the agent speaks at the start of the call — alongside the consent-gated recording disclosure when recording is enabled. Turning it off is an explicit, logged decision by the account, not an accident of omission. Defaults are the strongest compliance control a platform has, because most users never change them — a fact you can use in your favor.

The disclosure rule is one of four controls an outbound AI operation needs on day one — the others are calling-hour windows, Do-Not-Call screening, and recording consent. Vosy enforces all four in the dial path. Learn how the platform works →