Anthropic is updating Claude so its memory is shared across Chat and Cowork, allowing users to carry context between conversations and agent tasks without repeatedly briefing the AI.
Starting August 25, information Claude remembers from chats can also be used when running tasks in Cowork, and context learned through Cowork can carry back into chat. Anthropic said the change is intended to make the two experiences function more like a single continuous assistant.
Claude will also expose its stored memories directly to users. Memories are organized by topic and can be viewed, edited, or deleted individually. The system updates those entries as conversations happen rather than waiting until a conversation ends to create a summary.
Anthropic said Claude will avoid saving sensitive topics by default, including information related to health, race, ethnicity, religion, political beliefs and gender identity. Users can choose to allow sensitive topics through their memory settings and will receive a notice when Claude saves information it considers sensitive.
Some information will not be stored even when that setting is enabled, including Social Security numbers, government identification numbers, criminal history and immigration status.
