Claude’s memory system is becoming more user-editable. Engadget reports that Anthropic has been rolling out more granular controls for Claude’s memory, letting users manage the topics the chatbot retains from prior conversations and decide when it should avoid saving context at all. The underlying feature is not new. According to Engadget, Claude has had memory since summer 2025, allowing the chatbot to use earlier chats as context for later prompts. The practical aim is personalization: Claude can avoid asking users to repeat background information that has already come up in previous interactions. Engadget reports that memory is enabled by default for all Claude users, including free accounts. A recent update also lets Claude carry memory across both ordinary chats and Claude Cowork sessions, according to the report. That makes the accuracy of stored information more consequential, because a stale or incorrect memory can shape future answers across more of a user’s workflow. The main change is control at the topic level. Engadget says Claude organizes memory around specific topics, which users can edit from the Claude web client at claude.ai. The report frames this as the simplest place to review and adjust what Claude believes it should remember about a user. Anthropic has also added a setting called “Include sensitive topics in memory,” according to Engadget. By default, Claude will not save memories about sensitive areas such as health, race, ethnicity, religious beliefs, politics, or gender identity. If a user enables the option, Claude can begin saving some sensitive-topic memories, and Anthropic says users will see a notice when Claude generates one. Engadget notes an important limitation from Anthropic: users on an older version or the mobile app will not see that notice, and Claude will not save the sensitive topic to memory. There are limits even when the sensitive-topic setting is enabled. Engadget reports that Claude will not retroactively save past conversations, and that some information, including government ID numbers, will never be saved to memory. Users who do not want Claude to generate memories can turn off “Generate memory from chats,” according to the report. Anthropic then offers two options: pausing memory or resetting it. Engadget says Anthropic warns that resetting memory cannot be undone; if the user later re-enables the feature, Claude starts again without its previous memory. Claude’s project system adds another layer. Engadget reports that Anthropic silos memory in projects so Claude’s context is not overloaded with information unrelated to the current task. Users can shape project memory by adding comments, documents, and GitHub information through the interface, and Claude generates a daily summary of project work that users can edit. If a conversation starts outside the intended project, Engadget says users can move it into a project through the three-dot menu. For information a user wants Claude to remember across tools, Engadget points to the Profile section of the Memory menu. For moments when a user wants to avoid memory capture temporarily, Claude also has an Incognito mode accessible from the ghost icon in the interface, according to the report. Who benefits: Claude users who rely on the assistant for repeated work benefit if accurate context carries across chats, projects, and Claude Cowork sessions. Teams using projects may also benefit from more bounded memory around specific workstreams. Who's exposed: Users are exposed if Claude stores inaccurate or unwanted context and then applies it later. The sensitive-topic controls and Incognito mode matter most for people who mix personal, regulated, or work-sensitive information into chats.