Anthropic is asking Claude users to hand over their voice conversations for training

Audio recordings and voice chat data are the newest item on Anthropic’s list of things it would like Claude users to share. According to a report published October 4, 2026, a prompt now appears for people who open Claude’s voice features and asks permission to use their voice data to improve Anthropic’s AI models. It offers two buttons, Allow and Not now, and declining leaves everything as it was, because the voice setting is off by default and sharing happens only if a user switches it on.

The request is separate from the one Anthropic put to Claude users in 2025 about typed chats and coding sessions. That earlier choice covered text; this one covers recordings of a person speaking, which Anthropic describes as audio recordings and voice chat data, and it arrives as its own setting with its own default.

The Allow and Not now prompt

BleepingComputer, which reported the change, says Anthropic has started asking Claude users to voluntarily share their voice conversations to help train and improve its models. Its reporter, Mayank Parmar, quotes the explanation shown alongside the prompt: “Audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech.” The BleepingComputer report is dated 06:53 on October 4.

The report quotes the text of the prompt and the settings, not a spokesperson, and it does not attribute any further explanation to Anthropic beyond that wording. What it establishes is the mechanism: a notification that surfaces when someone uses Claude’s voice capabilities, an explicit request for permission, and a toggle in Privacy settings that starts in the off position. The report as read does not describe how long recordings would be kept.

The stated purpose is narrow. The wording ties voice data to speech, meaning how Anthropic’s models understand what a person says and respond to it, and the request is framed as a choice made at the moment someone uses voice, not as a change applied to an account in the background.

The Settings and Privacy toggle

A matching control lives under Settings, then Privacy. The report describes the voice option as disabled by default, so Anthropic is not automatically opting users into sharing their voice recordings, and quotes that wording. A user who agrees can reverse the decision later: the setting can be switched off at any time, and voice data that has already been shared can be deleted from the same settings area.

Voice sharing is separate from the choice Anthropic already offers for typed chats. The report says voice training operates independently of the existing option to let chats and coding sessions be used for model improvement, which means a user can allow one and decline the other. Someone who agreed to chat training in the past has not, on that description, agreed to anything about voice, and someone who declines the voice prompt has given up nothing about how typed chats are treated.

The chat training rules that already exist

Anthropic’s help center describes the earlier option in its article on whether data is used for model training. It says users choose whether to allow chats and coding sessions to be used to improve Claude, covers the consumer plans Free, Pro and Max, and sends commercial products such as Claude for Work and the API to a separate policy. Incognito chats are exempt even when model improvement is on. That article, as read for this report, is dated March 16, 2026 and does not mention voice or audio.

Anthropic announced the chat-training choice on August 28, 2025, in its post on updates to its consumer terms. That update extended retention to five years for users who consent and kept the existing 30-day period for those who decline, and it said the changes did not apply to Claude for Work, Claude for Government, Claude for Education, or API use through third parties. The post said users stay in control of that setting and of whether their data is used in this way, a statement that sits alongside the new voice prompt even though it was written for chats.

Anthropic’s data retention article says deleted conversations leave the back-end systems within 30 days and that data from users who enabled model improvement may be kept in a de-identified form for up to five years in training pipelines. Another article explains how to change the model improvement setting, and it too concerns text chats and coding sessions.

None of those pages gives a retention period for voice recordings, and the BleepingComputer report does not state one either. Anthropic’s pages for chats name five years for consenting users, and a period specific to audio is the detail the available material leaves unstated.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview