Anthropic has introduced a shared memory system for Claude chat and Claude Cowork, making the information a person builds up in one surface available in the other. The change is intended to reduce repetitive briefing when a user moves from a conversation into a cloud task, or returns from a task to chat. It is an important product update because it ties longer-running work to the context users have already chosen to retain.
Memory now works as a single layer across the two products. A person can discuss a project, priorities or preferred working style in chat, then ask Claude Cowork to carry out a related task without starting again from a blank slate. Details that emerge while a Cowork task is underway can also be available later in chat. Anthropic says this makes it easier to pick up work rather than repeatedly recreate the same context.
One context across two ways of working
The practical purpose is continuity. Someone preparing a conference in chat might discuss the city, speakers and agenda, then hand budgeting and logistics to Cowork with that background already available. A manager’s preferred update format or a team’s definition of a metric can similarly travel with later tasks. The feature does not turn every conversation into a permanent record by itself; it gives users a clearer mechanism for seeing what has been retained and changing it.
Anthropic says Claude updates memory as people chat, rather than relying solely on an end-of-conversation summary. A changed project deadline, for example, can be reflected in the next interaction. Users can pause memory or reset it. This makes the feature useful for ongoing work while preserving an escape hatch for someone who wants a fresh start or who is working on a temporary topic.
Visibility and editing are central
Every saved item is exposed as a short file under Topics in Claude’s Memory settings. Users can read, edit or delete those files. That design gives people a direct way to correct a stale company name, an outdated preference or a mistaken detail once, instead of having to correct it again in each new conversation. It also makes the retained context more inspectable than an opaque personalisation setting.
The emphasis on controls is notable. Memory can make an assistant more useful, but it creates a higher expectation that users can understand what information is shaping later responses. Topic-level editing gives a person a practical review surface, while pause and reset controls provide broader choices. The launch positions memory as something users direct, not merely a background feature that expands without visibility.
Sensitive topics remain off by default
Anthropic says Claude does not store topics that are considered personal or sensitive by default, including health, race, ethnicity, religious beliefs, politics and gender identity. Users who want context from these areas can enable an “include sensitive topics in memory” setting. When that option is on, Claude shows a notice each time it saves a sensitive topic, and it applies only going forward rather than retroactively capturing earlier conversations.
Some information remains excluded even if a user enables sensitive-topic memory. Anthropic identifies sensitive identification numbers, criminal history, immigration status and material that conflicts with its acceptable-use policy as examples. The distinction is significant: user control is expanded for certain personal context, but it is not presented as permission to retain every category of information. People should still consider what details are appropriate to share with an AI service for their own circumstances.
Availability differs by plan and organisation
Memory is on by default for Free, Pro and Max users on web, desktop and mobile, although storing sensitive topics is not. Mobile users may need the latest app version. For Team and Enterprise customers, administrators control whether the capability is available and it begins switched off for individual users until they choose to enable it. That split recognises that organisations often need a separate decision about how a feature is introduced and governed.
For workplace users, the release raises familiar operational questions: which teams should have access, how should training explain the settings, and what internal guidance applies to sensitive or confidential material? The product’s controls do not remove an organisation’s responsibility to set appropriate practices. They do, however, make it easier to align a conversational assistant with the context of sustained work when those practices are in place.
A more persistent Claude experience
The shared-memory launch makes Claude Cowork feel less like a separate task runner and more like an extension of a user’s ongoing interaction with Claude. The benefit is not simply fewer repeated prompts. It is the ability to move between thinking through a problem and asking for work to be done while maintaining the relevant context. For users who value continuity but also want inspectable, editable controls, Anthropic’s approach offers a more deliberate form of personalisation.