Anthropic is making a significant change to the way Claude’s memory, unifying the content it acquires from conversations with that it collects through Cowork. Now, the two experiences no longer operate as isolated environments, but share common knowledge for the user, creating the image of a single digital assistant.
A memory for the entire Claude ecosystem
With the new architecture, information generated in a regular conversation can be used later by Cowork and vice versa. Anthropic says the change applies to individual users of the Free, Pro and Max plans and applies to web, desktop and mobile.
The practical significance is great. Claude no longer needs to treat each conversation as a new beginning. If, for example, a user mentions that a work deadline has been moved to September, the information can be available in a later task, without requiring an explicit “remember this” command.
Memory is updated while you talk
The second major change concerns storage time. Previously, Claude waited for a conversation to finish before generating a summary and deciding what was worth keeping. Now, it can add new topics to memory during the interaction itself.
See also: Anthropic accuses Alibaba of the largest distillation campaign against Claude
This makes the experience more natural, but at the same time reduces the amount of time the user had to decide whether a conversation should ultimately become part of their long-term profile.
From brainstorming to automatic follow-up
Anthropic gives typical examples. A user can ask Cοwork to prepare a briefing for their boss and the system already knows who they are and what style they prefer.
Similarly, a discussion about organizing a conference can lead to future work where Cowork already knows details like the number of attendees, location, and speakers. Thus, Claude starts to function more as a unified assistant and less as a collection of different tools.
The user can check what Claude remembers
Transparency is a key part of the new approach. Stored information appears in the “ Topics ” section of the memory settings. There , the user can read, modify, or delete individual items , with the changes taking effect in subsequent interactions.
There are also toggles to enable or disable memory, as well as a separate control for sensitive information. By default, Claude does not store information about health, race, ethnicity, religious beliefs, politics, or gender identity.
The user can choose otherwise, but activation is not retroactive. In addition, Claude displays a notification when it stores information that falls into these categories.
See also: Anthropic: Claude uploaded malicious package to PyPI in security testing
What data is excluded?
There is information that remains out of memory even if the relevant options are enabled. This includes social security numbers, government identification documents, criminal records and immigration status information , as well as content that violates Anthropic's usage policies
This is an important security measure, especially since Cοwork is not limited to text production, but can perform tasks and interact with the computer environment.
Claude Code remains unique
The unification does not cover Claude Code, which remains essentially a separate “island” of memory. This shows that Anthropic’s effort to create a unified user interface is not complete.
Meanwhile, the Cowork Chrome extension further expands the scope of the functionality. Information from interactions within the browser can now be part of the broader user experience.
Why memory is becoming more important
This particular development has a different weight because Cοwork is an agent and not just a chatbot. A system that performs tasks and at the same time has months of accumulated knowledge about the user acquires a clearly greater operational scope.
See also: Anthropic launches Claude Science, an AI workbench for the lab
For businesses, Anthropic takes a different approach: in Team and Enterprise environments, memory is not automatically enabled for employees and control passes more to administrators and compliance teams.
The new battle of AI assistants
Memory is now becoming a key competitive advantage in the AI market. The more an assistant knows about someone's work style, preferences, and projects, the harder it becomes to transfer that experience to a different service.
That's why the Chat and Cowork integration is not just a technical upgrade. It's a step towards a Claude that functions as a stable, personal assistant. For users, the most important thing now is to regularly visit the "Memory" settings and check the list of information that has been saved.
