Anthropic asks Claude users to share conversations for AI training

Anthropic wants to hear from you. It means that literally.

The company behind the Claude chatbot is asking users to voluntarily share their voice conversations to help improve its AI models.

The request lands in a crowded debate over how AI companies collect training material. Anthropic has long marketed itself as the safety-focused lab. That makes every change to its data practices worth a close look.

What Anthropic is asking for

The core of the request is simple. Anthropic is prompting users to opt in to sharing voice conversations so the company can use them for model improvement.

That framing matters because of where Anthropic started. Under its earlier setup, the company’s voice features worked differently from its text chats. Audio recordings were deleted after transcription and were not used to train models.

Put simply, the spoken audio vanished once it became text. Only the transcript could feed training, and only if a user’s settings allowed it.

How the 2025 policy shift set the stage

The voice ask builds on a bigger change Anthropic made on August 28, 2025. That day, the company updated its consumer terms to let users consent to their chats and coding sessions being used for model improvement.

This was a real departure. Before the update, Anthropic did not train on consumer data at all.

The updated terms covered users on the Claude Free, Pro, and Max plans. New users saw a prompt at signup. Existing users were given an opt-in deadline listed as September 28/October 8, 2025.

One detail drew attention. The “Help improve Claude” toggle was switched on by default.

The retention numbers also changed substantially. Users who opted in could have their data kept for up to five years. Users who opted out kept the old arrangement, with data retained for about 30 days.

Anthropic attached several guardrails to the policy:

Only new or resumed conversations qualify for the extended retention. Old chats sitting untouched in your history were not swept in retroactively.

Deleted conversations are excluded from future training runs. Deleting a chat effectively pulls it out of the pool.

Privacy settings can be changed at any time. A user who opts in today can reverse course later.

The changes also stopped at the consumer line. Anthropic’s commercial products, including its API and enterprise offerings, were not covered by the new terms.

Why the consumer versus enterprise split matters

That carve-out tells you a lot about how AI companies think about trust. Businesses paying for API access or enterprise contracts typically care intensely about where their data goes. Training on it would be a fast way to lose those customers.

Consumers, meanwhile, are where the volume lives. Anthropic has framed real user interactions as a way to improve both capability and safety, giving its models exposure to how people actually use them rather than how researchers imagine they will.

What this means for users and the AI race

For users, the calculation is fairly direct. Sharing voice data may help Claude get better at understanding speech. It also means handing over a more personal form of data than typed text. Anyone weighing the prompt should check their privacy settings and remember that the choice can be revisited.

Investors and observers tracking the AI sector may want to monitor how opt-in rates and user sentiment differ across Free, Pro, and Max plans, and whether enterprise customers remain comfortable with the separation.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.