Anthropic has unveiled a major update to the memory system of its AI assistant Claude, unifying the context between regular chats and the Claude Cowork work tool into a single, cohesive structure. Prior to this update, users regularly found themselves in a situation where, after a detailed discussion of a project in chat, they had to re-explain all the key details — from deadlines and participants to technical requirements — for the Cowork tool to correctly perform a specific task. Now Claude functions as a single assistant that "remembers" everything discussed in any session and instantly applies that knowledge when moving to practical work.
Unified Memory Instead of Fragmented Context
The practical effect of the update can be illustrated with a simple example: if during a conversation the user discussed with Claude the details of an upcoming conference — the number of participants, the host city, the list of speakers — then when subsequently creating an agenda or a final report in Cowork, the AI will automatically take all of this data into account without additional prompts or repeated requests. This eliminates one of the main irritants when working with AI assistants: the need to constantly "remind" the model of previously agreed-upon project parameters.
Real-Time Memory Updates
The key technical change is that Claude no longer waits for the end of a conversation to summarize and save information to memory. New topics, facts, and decisions are captured directly during the interaction, making the transition between chat and Cowork virtually instantaneous. The user does not need to specifically ask the model to "remember" a particular parameter — the system does this automatically and continuously, updating the context as the dialogue develops.
Full User Control Over Saved Data
The developers have also given users direct access to the saved memory lists. At any time, you can view, edit, or completely remove a mention of any topic that Claude has recorded. This solves the "black box" problem, where the user did not understand exactly which data the model had saved and how it influenced subsequent responses. Now, memory transparency is part of the user interface.
Privacy Boundaries: What Claude Will Never Save
By default, Claude does not store personal or sensitive information: data about health status, racial or ethnic origin, religious beliefs, political views, or gender identity. Users who consciously want such topics to be considered in the context can enable the corresponding "include sensitive topics in memory" toggle in the settings. In this case, the app will send a notification each time a sensitive topic is recorded in memory. At the same time, Anthropic has established an absolute ban on storing official documents — passport data, social security numbers — as well as information about criminal records and immigration status. These categories of data will not be saved under any circumstances, regardless of the user's settings.
Availability and Platforms
The updated memory feature is enabled by default for users on the Free, Pro, and Max plans. It is available in the web version, the desktop app, and on mobile devices running iOS and Android. To work on mobile platforms, the app must be updated to the latest version. Thus, Anthropic has made unified memory available to its entire audience, including free users, underscoring the strategic importance of this feature for Claude's competitiveness in the AI assistant market.
