Eigen RadarAI
Archive
Analysis

Claude asks for separate permission to train on voice recordings

Anthropic is asking Claude users for permission to use recordings and voice-chat data in model training. The choice is optional and disabled by default in the interface examined. It operates separately from permission to use text chats and Claude Code sessions. Users are offered controls to disable sharing or delete voice data later; an account-wide rollout schedule has not been established.

Artificial Intelligence··Evening
Seen from behind, a silver-haired woman in a lime raincoat holds a phone near her ear on a sunlit porch.

A new permission prompt in voice conversations

Anthropic is asking Claude users to voluntarily share voice conversations for model training. The request appears in the product's voice features and asks for permission to use spoken interactions to improve its models. It introduces a choice about sharing voice data rather than announcing a new speech model or a measured improvement in Claude's performance.[1], [2]

BleepingComputer reporter Mayank Parmar describes a prompt covering audio recordings and voice-chat data. Users can accept the request or choose to postpone it. In the interface the publication examined, sharing was optional and the associated privacy switch was disabled by default. Those observations describe the version that was inspected. They do not establish that the same interface has reached every Claude account, that all product versions expose identical controls, or that an account-wide launch has been completed. An account-wide rollout timetable has not been announced.[1]

Voice and text permissions are separate

The control appears under Settings and Privacy. Its description refers to approved sessions and specifies recordings alongside voice-chat data. Voice permission is separate from the existing option for using ordinary chats and coding sessions in training. A user can allow voice conversations while excluding text chats and Claude Code sessions, or choose the reverse. The interface does not require agreement to one category as a condition of agreeing to the other. Users manage these choices separately. How many have enabled either setting, and how much data Anthropic has collected through them, has not been established.[1]

Controls offer disabling and deletion

The prompt says users can later disable voice sharing or delete the data from settings. This is a promise displayed in the interface, rather than an independent measurement of deletion behavior. The retention period for recordings and the effect on training completed before a user changes the setting remain unspecified. It also does not establish a universal default beyond the inspected interface. The described material comprises recordings and voice-chat data; the reported purpose is to improve speech understanding and responses. No performance results accompany the additional consent control.[1]

References

  1. News sourceBleepingComputerClaude asks for separate permission to train on voice data↩1↩2↩3↩4
  2. News sourceTestingCatalogAnthropic starts prompting users to voluntarily share their voice conversations for training purposes.↩
AI safety and regulationChip wars and the supply chainThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →