Anthropic has started asking Claude users whether it can use their voice conversations to train and improve its AI models, BleepingComputer reported Sunday. The request appears when people use Claude's voice features, and a new switch in the app's privacy settings controls the same choice. In the version BleepingComputer saw, the setting is off unless a user turns it on.

What Happened

The prompt asks users to allow Anthropic to use their voice data to improve its models and offers two buttons, Allow and Not now. Anthropic says audio recordings and voice-chat data help its models get better at understanding and responding to speech.

The same option sits under Settings, then Privacy, where it is labeled as permission to use voice data. Its description says turning it on allows Anthropic to use a user's audio recordings and voice-chat data from approved sessions to improve its AI models.

RelatedCharacter.AI Maker Files for Chapter 11 With Google Its Largest Unsecured Creditor

According to BleepingComputer, the setting is disabled by default, so users are not enrolled automatically. It is also separate from the existing setting that governs whether Anthropic can train on text chats and Claude Code sessions. A user can allow one and decline the other.

Anthropic says users who agree can switch the setting off later and delete their voice data from the settings menu.

Why It Matters

Voice assistants depend on large amounts of real speech to handle accents, background noise, interruptions and the informal way people talk compared with how they type. Recordings from actual product use are among the most valuable data for that purpose, and among the most sensitive, because a voice is biometric and audio can capture other people in the room.

By making the voice choice opt-in and keeping it apart from text training, Anthropic is giving users a narrower decision than a single all-or-nothing toggle. Someone who is comfortable sharing typed conversations may feel differently about recordings of their voice, and the split lets them act on that.

The approach also creates a clear record of consent for each type of data, which matters as regulators in Europe and U.S. states pay closer attention to how AI companies use consumer conversations for training.

Open Questions

BleepingComputer's report leaves several details unanswered. It does not say how long Anthropic keeps voice recordings once a user opts in, how quickly deletion takes effect, or whether recordings already made before a user agrees are covered. It also does not say how many users have seen the prompt or whether it is appearing across all plans and regions.

Another open point is what counts as an approved session. The setting's description refers to voice data from approved sessions, but the report does not explain whether that means every voice conversation after opting in or only some of them.

What to Watch

Anthropic's privacy policy and help pages are the likeliest places for retention periods and other specifics to appear. Also worth watching is whether the prompt reaches users in the European Union, where consent rules for voice data are stricter, and whether Anthropic later changes the default or the way the prompt is presented.