Since summer 2025, Claude has had the ability to remember past conversations, allowing Anthropic's chatbot to personalize responses and reduce context repetition. By default, memory is enabled for all users, including free accounts, and following a recent update, the feature carries over across chats and Claude Cowork sessions. Anthropic has rolled out more granular controls, allowing users to edit specific topics the AI retains.

To edit memories on the web client, users must select Settings from the profile menu, click Memory, and use the Edit icon next to specific Topics. On mobile, tapping the profile icon accesses Settings, then Capabilities, then Memory files, where topics can be selected and modified via the prompt bar or deleted using the trash icon. A new option, Include sensitive topics in memory, allows Claude to record health, race, ethnicity, religious beliefs, politics, or gender identity data, though this is disabled by default.

Users will see a notice each time a sensitive memory is generated. The chatbot will never save government ID numbers, even with this option enabled. Users can disable memory generation entirely via the Generate memory toggle, with options to Pause or Reset memory; resetting cannot be undone and will start Claude from scratch.

Additionally, Claude silos memory within projects to prevent context window overload, and users can add specific information to the Profile section for cross-tool recall. To avoid memory writing, users can switch to Incognito mode via the ghost icon, which ensures chats are not saved or used for training.