What happened
TechCrunch reports that Anthropic has merged the memory systems used by Claude chat and Claude Cowork. According to the report, Claude can now carry relevant information from conversations into Cowork, while users can inspect, edit, or delete retained topics. The memory feature is reported to be enabled by default for Free, Pro, and Max users across web, desktop, and mobile, with an app update required on iOS and Android.
TechCrunch reports that Anthropic is merging the memory system used by Claude’s regular chat experience with the one used by Claude Cowork. Cowork is presented in the report as the part of Claude used to take action after a user has discussed or researched a project. Before the change, TechCrunch says, information developed in chat did not automatically carry over when a user switched to Cowork, requiring the user to repeat details. The reported change is therefore a product integration, not a new underlying model announcement.
According to TechCrunch, the shared memory is intended to let Claude use context from earlier conversations when users move between discussion and execution. The report gives examples involving a manager update or a conference agenda. In those cases, Claude could already understand references such as a conference’s headcount, city, and speakers, rather than requiring those details to be entered again in Cowork. TechCrunch characterizes the result as making Claude feel more like one continuous assistant instead of two separate products.
TechCrunch also reports that Claude will add topics to memory during an ongoing chat instead of waiting to summarize the conversation after it ends. The report says this should make memories update and become available between chat and Cowork more quickly. Anthropic is also reported to be exposing retained information to users, who can read, edit, or delete memories on any topic. The source does not provide independent testing of whether those controls remove all relevant information from future use or how quickly changes take effect.
The report says the feature is enabled by default for Claude Free, Pro, and Max plans on the web, desktop, and mobile. On iOS and Android, TechCrunch says users must update to the latest app version. These availability details come from the report and are not independently confirmed in the supplied source. TechCrunch does not provide a detailed rollout schedule, regional availability map, or technical explanation of whether the same memory store is used across every device and plan.
Read the primary source: techcrunch.com ↗
Why it matters
The change addresses a practical limitation of AI assistants: users previously had to repeat context when moving from conversation and research into task execution. A shared memory could make Claude feel more like one continuing assistant, but it also increases the importance of accuracy, transparency, privacy controls, and user understanding of what is retained.
Shared memory can reduce one of the most visible sources of friction in AI-assisted work: repeatedly re-entering background information. TechCrunch’s examples point to a workflow in which a user first explores an idea in chat and later asks Cowork to produce or change something. If the system reliably carries over relevant context, users may spend less time reconstructing project details and more time checking the resulting work. The practical benefit depends on memory being accurate and appropriately scoped.
The change also shifts the user’s relationship with an AI assistant from a sequence of mostly separate conversations toward a continuing record of preferences, projects, and facts. That continuity can be useful, but it means a mistaken memory may influence later outputs without being obvious to the user. A retained detail could become outdated, be misunderstood, or be applied to the wrong task. The ability to inspect and edit memories is therefore central to the feature, rather than a minor convenience.
Privacy is a particularly important part of the reported change because the system can retain information across contexts. TechCrunch reports that Claude will not store personal or sensitive information by default, including health data, race, ethnicity, religious beliefs, political views, and gender identity. The report says users can opt in to including sensitive topics and will be notified when such a topic is saved. It also says certain information, including government-issued identification, Social Security numbers, criminal history, and immigration status, will never be saved under Anthropic’s stated rules.
The supplied report does not independently verify Anthropic’s implementation or test its privacy boundaries. It also does not explain whether memories are used to improve models, how deletion interacts with backups or logs, what happens when users share an account or device, or how the system distinguishes a temporary statement from a durable preference. Those unknowns limit how confidently the feature’s privacy implications can be assessed. The most defensible conclusion from the source is that Anthropic is making memory more integrated and more visible, while many operational details remain unspecified.
What to watch next
The report does not independently establish how consistently the feature works, how quickly memories update in practice, how long information is retained, or how the system handles errors and conflicting instructions. Users should watch for Anthropic’s documentation on retention, deletion, data use, regional availability, and safeguards around sensitive information.
The first issue to watch is reliability. TechCrunch reports the intended behavior, but the supplied source contains no independent hands-on evaluation of whether Claude consistently retrieves the right memory, ignores irrelevant information, or handles contradictions between old and new details. Useful follow-up reporting would test ordinary workflows across chat and Cowork, including corrections, project changes, and requests involving multiple unrelated topics.
Users should also look for clearer information about retention and deletion. The report says people can read, edit, or delete stored information, but it does not state how long memories remain available, whether deletion is immediate across devices, or whether copies persist in logs, backups, or other systems. Anthropic’s documentation would need to clarify these boundaries for users to understand what deleting a memory actually accomplishes.
Sensitive-information handling deserves continued scrutiny. TechCrunch reports that sensitive topics are excluded by default, can be enabled by a user preference, and trigger a notification when saved. It does not describe the full list of categories, how sensitive information is detected, what happens when a user discusses someone else, or whether the controls work identically on web, desktop, and mobile. Independent testing could establish whether the stated safeguards match observed behavior.
Finally, watch for the feature’s effect on work performed by Cowork. The source describes smoother movement from research into action, but provides no measurements of time saved, accuracy, task completion, or error rates. Users and organizations should treat continuity as a convenience that still requires review, especially when remembered information affects documents, plans, or other consequential work. The report also leaves open whether the rollout is universal or subject to regional, account, or platform limitations.
