Anthropic gives chat and Cowork one memory

Anthropic recently updated its Claude AI to provide a unified memory experience across its chat and Cowork services. Previously, Claude's memory in Cowork was separate from its chat memory, but this new system allows Claude to remember information from both services in a single, cohesive memory.

With this update, Claude will be able to remember important facts and context, such as user preferences and project details, regardless of whether the user interacts with Claude through chat or Cowork. This is particularly useful in Cowork, where users often engage in in-depth work that requires recalling prior conversations.

An example provided by Anthropic illustrates the utility of this feature: if a user asks Claude to draft an update for their manager, the AI will already have the necessary context and know how the user likes updates written. Similarly, if a user brainstorming ideas for a conference in chat, Cowork can then use that information to generate a budget and logistics document that includes relevant details like headcount, city, and speakers.

Another notable improvement is that Claude now updates its memory in real-time while users are working with it, rather than only after the chat ends. This means that the context built up over months of conversations, such as Q3 priorities and project statuses, is readily available when Claude is asked to perform a task. The AI will also carry over any new information generated in Cowork back to chat, ensuring a seamless user experience.

However, Anthropic has also implemented certain safeguards to protect user privacy. By default, Claude will not store any sensitive information in its memory, including health data, religious beliefs, race, ethnicity, and gender identity. Users can choose to include sensitive topics in the settings menu, but there is currently no option to selectively include or exclude certain topics.

Anthropic has also explicitly stated that Claude will not store sensitive identification numbers, criminal history, immigration status, or any content that violates their Acceptable Use Policy. Additionally, Claude remains ad-free, meaning that no stored information in memory is used for targeted advertising, similar to OpenAI's approach with their Computer History feature.