In brief

People should have a simple, nationwide way to refuse unsolicited calls generated or conducted by AI. Once a person opts out, businesses and campaigners should not be able to call through an artificial voice or autonomous conversational system merely because the telephone number remains technically reachable.

Every permitted AI-generated call should also identify itself clearly at the beginning. Disclosure and the right to opt out solve different problems: knowing what is calling does not mean consenting to receive it.

Context

Automated calling can be produced at far greater scale and lower marginal cost than human calling. Generative voices make such calls more conversational and potentially harder to recognise. Without a practical refusal mechanism, the cost of that scale is transferred to everyone whose attention, time, or trust is targeted.

The proposal should distinguish unsolicited outreach from a call expressly requested by the recipient, such as an accessibility service, appointment reminder, delivery coordination, or account alert.

The blueprint

Create a national preference that allows a person to reject AI-generated calls across providers. The preference should be easy to set, easy to verify, and independent of any one telecom operator or calling business.

Consent for a particular organisation or purpose should be specific and revocable. It should not be buried inside general terms or inferred from an unrelated transaction.

Implementation

Calling organisations would be responsible for checking the preference before initiating an AI-generated call, recording the legal basis for permitted calls, and transmitting a trustworthy machine-readable indicator that the call is automated.

Telecom networks could use that indicator to label or block calls according to the recipient’s preference. Enforcement should reach both the organisation commissioning the call and the technical provider placing it, while protecting providers that follow verified instructions in good faith.

Trade-offs

Bad actors may conceal or falsify the automated-call indicator. A blanket rule could also block useful services chosen by the recipient. The design therefore needs authenticated call metadata, narrow exceptions, purpose-specific consent, and a route for legitimate callers to correct classification errors.

Open questions

  • Should the preference extend to all automated calls or only generative conversations?
  • How should requested reminders and public-interest alerts be treated?
  • Can existing unwanted-communication systems carry the preference?
  • What penalties would deter high-volume violations across jurisdictions?