BleepingComputer reported on 4 October 2026 that Anthropic has begun asking Claude users to share their voice conversations for AI training. The prompt shows up when someone uses Claude's voice features. Users can pick Allow or Not now. BleepingComputer says a matching switch sits in Claude's Privacy settings. It also says this choice is separate from Anthropic's existing option to train on regular chats and coding sessions. A user could allow one and refuse the other.
BleepingComputer says that in the interface it saw, the setting is off by default. So users are not signed up automatically. The prompt text says people can turn the option off or delete the data later in settings. The settings page says it applies to "approved sessions". The report does not explain what that phrase covers. It also does not say when the change started, which users or regions see it, or how long Anthropic keeps the data. The report quotes no Anthropic statement beyond the text in the app.
For teams that buy or run Claude, this adds one more privacy switch to check. It is separate from the chat and Claude Code training setting. Anyone setting data rules for staff who use voice features may need to review both.