Anthropic asks Claude users to opt in to sharing voice chats for training, report says
In brief
- Anthropic asked Claude users to voluntarily share voice conversations for model improvement, Crypto Briefing reported.
- Voice audio was previously deleted after transcription and not used for training, per the report.
- The request builds on Anthropic's August 28, 2025 consumer terms update covering chats and coding sessions.
- Opted-in data could be retained up to five years; opt-outs kept roughly 30-day retention.
- API and enterprise products weren't covered by the updated consumer terms.
What changes for voice
Anthropic prompted users to opt in to sharing voice conversations so the company could use them for model improvement, Crypto Briefing reported.
The request is framed as voluntary.
That's a shift from the policy the outlet described. Under the earlier setup, audio recordings were deleted after transcription and weren't used for training. Only the transcript of a voice conversation could feed training (and only if a user's settings allowed it).
The August 2025 backdrop
The voice ask isn't the first change of this kind. On August 28, 2025, Anthropic updated its consumer terms to let users consent to their chats and coding sessions being used for model improvement. Before that update, Anthropic didn't train on consumer data, according to the report.
Those terms covered users on the Claude Free, Pro and Max plans. New users saw a prompt at signup, and the "Help improve Claude" toggle was switched on by default.
Retention is where the two paths split. Users who opted in could have their data kept for up to five years, while users who opted out kept the prior arrangement of about 30 days. Only new or resumed conversations qualified for the extended window; old, untouched chats weren't swept in retroactively. Deleted conversations were excluded from future training runs, and users could change their privacy settings at any time.
Anthropic's commercial products, including its API and enterprise offerings, weren't covered by the new terms.
The safety pitch
Anthropic has framed real user interactions as a way to improve both model capability and safety, Crypto Briefing reported. It's an argument that fits the company's positioning.
"Anthropic has long marketed itself as the safety-focused lab."
The report doesn't give a launch date, plan coverage or retention terms specific to the voice request, so it isn't clear whether the August rules carry over to audio.
Frequently asked questions
Did Anthropic previously use Claude voice recordings to train its models?
According to Crypto Briefing's account of Anthropic's earlier setup, voice audio recordings were deleted after transcription and were not used to train models. Only the transcript could feed training, and only if a user's settings allowed it.
How long can Anthropic keep Claude chats used for training?
Under the consumer terms Anthropic updated on August 28, 2025, users who opted in could have their data retained for up to five years, Crypto Briefing reported. Users who opted out kept the prior arrangement of about 30 days, and deleted conversations were excluded from future training runs.
Do Anthropic's updated consumer terms apply to its API?
No. According to Crypto Briefing, Anthropic's commercial products, including its API and enterprise offerings, were not covered by the updated consumer terms. Those terms applied to users on the Claude Free, Pro and Max plans.
