Anthropic wants Claude users’ voice chats for training

Anthropic is asking Claude users to let it train and improve its models on their voice conversations, BleepingComputer reported on October 4. Sharing is voluntary.
The request extends a policy Anthropic reversed last year. Until then the company did not train on consumer chat data, TechCrunch reported in August 2025, and it then gave every Claude user until September 28 to decide whether their conversations could be used. The dates differ: Anthropic’s own page gave existing users until October 8, 2025. After that, they had to choose a setting before they could keep using Claude.
Those consumer terms cover Claude Free, Pro and Max, plus Claude Code when it runs from accounts on those plans. They do not cover Claude for Work, Claude for Government, Claude for Education or the API, including access through Amazon Bedrock and Google Cloud’s Vertex AI. Users who opt in already have their data kept for five years, and so does any feedback they submit on Claude’s responses. Users who opt out stay on the 30-day retention period. The setting applies only to new or resumed chats and coding sessions, and a deleted conversation is not used for future training. It is unclear whether voice recordings fall under the same five-year retention.
Anthropic says shared data helps “improve model safety, making our systems for detecting harmful content more accurate and less likely to flag harmless conversations,” and helps future models get better at coding, analysis and reasoning. It also says it filters or obscures sensitive data with a mix of tools and automated processes, and does not sell user data to third parties.
The chat-training setting is under Settings > Privacy > Privacy Settings, labelled “Help improve Claude”. A change there applies only to future sessions. Reopening an old chat makes it a resumed session, which the current setting then covers.
Up next
Claude asks permission to train on your voice chats
Per Testing Catalog, Claude started showing a prompt that asks to use your audio recordings and voice chats to train Anthropic’s models, though you can switch it off or delete that data anytime in settings.
Related stories
- Anthropic paid $740,000 to a firm led by Rubio’s ex-aide
- WSJ: Anthropic on Trump’s list of AI national champions
- Hong Kong and mainland funds barred from Anthropic IPO
- Four older Claude models get a watermark on November 16
- Anthropic joins a shared robots.txt opt-out for AI training
- Anthropic asks Australia to allow AI training with opt-out
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

