The news
Anthropic is asking Claude users to opt in to sharing their voice conversations so the company can use them to train its AI models, BleepingComputer reported on October 4, 2026. The prompt gives users two choices, "Allow" or "Not now," according to the report.
The voice-data setting is optional and disabled by default in Claude's privacy settings, BleepingComputer reported. Users can switch it off later or delete their voice data from settings at any time, according to the in-app text quoted by the outlet.
The in-app notice, as quoted by BleepingComputer, says audio recordings and voice chat data help Anthropic's models better understand and respond to speech. The outlet reported that the voice toggle sits apart from the existing option that governs whether text chats and coding sessions are used for training.
BleepingComputer did not say which Claude plans the voice prompt reaches, how long contributed voice data is kept, or whether business accounts are affected. Those details matter, because Anthropic's rules for text data differ sharply by plan and by setting.
Under consumer terms Anthropic published on August 28, 2025, the company began asking users of Claude Free, Pro and Max, including Claude Code used from those accounts, whether their chats and coding sessions could be used for training. Users who allowed it would have new data retained for five years, against 30 days for those who declined, the company said at the time. Anthropic said those terms did not apply to Claude for Work, Claude for Government, Claude for Education or API use, including through Amazon Bedrock and Google Cloud's Vertex AI.
The numbers
- Voice training default
- Off (opt-in)
- Text data retention if training allowed (2025 terms)
- 5 years
- Text data retention if declined (2025 terms)
- 30 days
Why CEOs should care
For most companies, the risk is not the enterprise contract but the personal account. Employees who use Claude Free, Pro or Max for work, and talk to it by voice, may say things they would never type: client names, deal terms, product plans. If they tap "Allow," that audio can be used for training. Because the voice switch is separate from the text switch, someone who already declined text training could still opt in to voice.
CISOs and compliance leads should update AI acceptable-use policies to cover voice mode explicitly. Spell out whether staff may use consumer AI accounts for company work at all, and if so, require training settings for both text and voice to stay off. Where possible, steer work use to business plans, which Anthropic's 2025 terms excluded from consumer training rules, and confirm with Anthropic whether the same exclusion covers voice.
General counsel should check whether recorded voice raises extra obligations. Spoken conversations can capture other people in the room, and some states treat voice and biometric data more strictly than text. Ask Anthropic, in writing, how long opted-in voice data is retained and whether deleting it removes it from future training.
The bigger picture
AI assistants are moving from typing to talking, and training data is following. Voice is a richer signal than text, carrying tone and accent along with words, which makes it valuable for building speech features and more sensitive to hand over.
Anthropic's approach here is opt-in with the setting off by default, as BleepingComputer described it. That is a different default from the 2025 change for text chats, when users were asked to make a choice to keep using Claude. Buyers should expect other assistant makers to seek voice data too, and should review each vendor's settings rather than assume they match.
What’s next
Watch for Anthropic to publish the retention period and plan coverage for voice data, and for any update to its privacy policy or consumer terms that spells out how voice training works for business users.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
