Anthropic on Tuesday merged the memory systems behind its Claude chatbot and Cowork, the agentic assistant it builds for everything other than software development, so that whatever one of them learns about a user becomes immediately available to the other. The change is live now and switched on by default for Free, Pro and Max subscribers across web, desktop and mobile.
The practical effect is the disappearance of a chore that has quietly irritated anyone using AI agents seriously: the rebrief. Someone who spent weeks refining a project in a chat window previously had to restate every detail when they moved to Cowork to actually produce the deliverable. The two products now behave like one continuous assistant rather than two tenants sharing a roof.
What the shared memory does
Anthropic frames the benefit as context that no longer has to be typed twice. Ask Cowork to draft an update for your manager and it already knows who that manager is and how they prefer updates written. Brainstorm a conference agenda in chat, and Cowork can carry the headcount, host city and speaker list into the budget and logistics documents it assembles afterward.
How memories get created has changed too. The system previously summarised a conversation once it had ended; topics are now written to memory continuously as the exchange unfolds. That matters for the handoff case, because a user can switch to Cowork mid-conversation and find the context already waiting rather than stale by a session.
The company is also opening up the store itself. Users can browse what Claude has retained topic by topic and edit or delete individual entries, rather than treating memory as an opaque blob that grows unattended.
Sensitive topics and the opt-in switch
By default the system withholds categories Anthropic classifies as sensitive, including health information, race and ethnicity, religious belief, political affiliation and gender identity. Users tired of re-explaining those details can turn on a setting that allows sensitive topics into memory, and the app notifies them each time one is saved.
Some categories stay permanently off-limits. According to Anthropic, Claude will never commit government-issued identification numbers, Social Security numbers, criminal history or immigration status to memory, regardless of what a user toggles.
No way to keep the two apart
The consolidation carries a constraint that privacy-minded users should register. There is no setting to separate what Claude Chat remembers from what Cowork remembers. Anthropic confirmed to The Register that the two products now draw on one shared system, and that maintaining a genuine wall between them requires running separate accounts.
The remaining controls are blunter instruments. Users can pause memory generation outright, route conversations they would rather not have retained through incognito chats, or manually prune topics after the fact. Team and Enterprise customers do not get to decide individually at all, since administrators control whether memory is available across the workspace.
Claude Code is the exception. Anthropic said the coding agent keeps its memory separate for now, and declined to say whether it will eventually join the shared pool.
Outlook
The move reflects a wider repositioning of memory from chatbot novelty into agent infrastructure. An assistant expected to execute multi-step work needs durable knowledge of who it works for, and rebuilding that picture every session is pure overhead. The open question is governance rather than capability: as memory becomes the connective tissue across a vendor's entire product line, how finely users can control it starts to matter as much as how much the assistant can recall.
