Anthropic said on August 25 that Claude’s memory now works the same way everywhere, ending the split between what the assistant remembered in chat and what it retained inside Claude Cowork, its agentic work platform.
What changed
Previously, memory in chat and memory in Cowork operated as separate systems, so a fact explained in one context often had to be repeated in the other. Anthropic says the two now draw on a single, shared memory that updates in real time as a conversation happens, rather than being summarized only after a chat ends. A cloud task in Cowork can now pick up context a user gave Claude in ordinary chat, and vice versa. In a company example, a user who explains once how their team defines its metrics no longer has to rebrief Claude before each quarterly report Cowork generates.
Users can see and edit what’s stored
The update adds a Topics view under Settings > Memory that lists everything Claude has retained, organized by subject, with options to read, edit or delete individual entries. Editing an item, such as correcting a company name, applies the fix across future conversations in both chat and Cowork.
How sensitive data is handled
By default, Claude excludes sensitive categories — health, beliefs, race, ethnicity, religion, gender identity and politics — from memory. Users can opt in via an “Include sensitive topics” toggle, which triggers a notification each time such information is saved. Certain categories are excluded regardless of that setting: Social Security and other government ID numbers, criminal history, immigration status, and anything that would violate Anthropic’s Acceptable Use Policy.
Availability
The unified memory is on by default for Free, Pro and Max individual plans across web, desktop and mobile, though phone users need the latest Claude app. Team and Enterprise organizations keep memory off by default, with admins controlling whether to turn it on.