Claude users can switch off model training and voice-data use, limit location access, and manage stored memories through settings. CNET’s guide also makes one important distinction: turning training off now cannot undo training that may already have used earlier conversations.
Anthropic Watch analysis
What happened
CNET outlines privacy controls for Claude on the web, iOS and Android. Under Settings, then Privacy, users can switch off “Help improve our AI models” and “Allow us to use your voice data”. Some accounts, including Enterprise plans, may not show those options because the data is not automatically collected for them.
On the web, location metadata can be switched off under Profile, Settings, Privacy. Mobile users can revoke Claude’s location permission through their device settings; CNET also describes an Android app setting. To review stored memories, open Settings and “Your Claude”, then “Your memory”. Users can edit or delete memories there, or turn off “Generate memory from chats”. CNET says incognito chats are not saved in chat history or used to train models. Its privacy guide has the full steps.
Why it matters
These controls affect different things: model improvement, voice data, location metadata and memory are not one big privacy switch. If you previously shared a sensitive conversation while training was enabled, CNET says deleting it and changing the setting cannot reverse any training that has already happened. Incognito mode is a forward-looking choice, not a time machine.
Location settings also have a trade-off. Turning access off can affect features that need your location, while Anthropic says it may still use an IP address to infer a coarse location for terms compliance and abuse prevention.
Our read
Start with the training and voice-data toggles, then check location permissions and “Your Claude” memory settings. Use incognito for sensitive chats going forward. The useful point is not that every concern has a neat off-switch, but that the controls are specific enough to check yourself. CNET’s guide is a practical map, not a promise that every account sees identical options.
What to watch
- Whether the settings differ by plan, device or account.
- How Anthropic explains the distinction between app location permission and IP-based location inference.
- Whether Claude’s controls change, especially for memory and data use.
Discussion spark: Should AI services make privacy controls easier to find even if that means more settings up front, or is a short default setup worth the trade-off?
Sources and evidence
- Source update (20 November 2026, 19:26 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.