Claude’s memory no longer stops at the chat window. Anthropic on Aug. 25 extended the same memory used in Claude Chat to Cowork, allowing information learned in one experience to carry into the other.
Users can or erase Claude’s memory, edit individual memories and keep sensitive topics excluded. They cannot, however, keep memory active in Chat while separately preventing Cowork from using it.
Claude memory now crosses into Cowork
In its Aug. 25 memory announcement, Anthropic said information remembered from chats can inform Cowork tasks, while context generated during Cowork work can carry back into Chat. Claude now saves individual memory topics as conversations happen rather than relying only on summaries created afterward.
Those memories can include projects, working preferences and people mentioned in conversations. The change extends Anthropic’s broader push to place Claude inside everyday workplace tools, including Claude agents operating inside Slack with access to approved conversations, tools and organizational data.
Memory is enabled by default for Free, Pro and Max users. Claude Code remains separate from the Chat-Cowork memory system for now, Anthropic told The Register, which also reported that there is no setting to keep Chat and Cowork memories separate on the same account while memory remains enabled.
Persistent memory can surface information in later tasks even when a user did not deliberately share the original conversation. In July, shared Claude chats appeared in Google Search, showing how information can move beyond the context users expected.
, reset or narrow what Claude remembers
Open Settings > Memory and switch off Generate memory from chats. Claude then gives users two choices:
** memory:** Existing memories remain stored, but Claude stops using them and creating new ones.Reset memory: Claude permanently deletes stored memories, including project memories. The action cannot be undone.
Users who want more selective control can open Settings > Memory > Topics to edit or delete individual memories. Include sensitive topics in memory is off by default, preventing Claude from saving categories such as health, race, ethnicity, religious beliefs, politics and gender identity unless the user opts in.
Saved memory is separate from Search and reference chats, which can retrieve information from previous conversations and has its own setting. Users trying to reduce how much historical context Claude can access should review both controls.
For conversations that should remain outside Claude’s ongoing context, Incognito chats do not use existing memory or add new memories. Anthropic still retains those chats for 30 days by default, and Enterprise retention policies can extend that period. Persistent context also adds another variable when agents can act through connected systems. A recent Claude Opus 4.6 API experiment showed how weak backend controls can let an agent operate beyond restrictions presented in an interface, making both access and retained context relevant when organizations decide where agents belong.
Read more: For a broader look at enterprise oversight, a Claude Code espionage campaign exposed governance gaps around AI agents connected to tools and operational data.