Claude now shares one memory across chat and Cowork, with full user control

From 25 August 2026, Claude has one memory, not two. Anthropic has unified the memory store that previously existed separately in Claude Chat and Claude Cowork into a single system that works across both surfaces. The change sounds simple, but it removes a genuine friction point that anyone who has used both products will have felt.

What changed, and why it matters

Until now, what Claude remembered about you in a chat conversation stayed in chat. When you handed a task off to Cowork, the agent started fresh. You might have spent a conversation refining a project brief, establishing your preferred tone, or explaining your company’s internal structure, only to repeat all of it the moment Cowork needed to do something with it.

That is no longer the case. As of 25 August 2026, the memory Claude builds during your conversations is the same memory Cowork draws on when running tasks in the background. Information flows in both directions: Cowork picks up what chat has learned, and anything Cowork learns comes back into chat.

Anthropic gives a practical illustration. If you have discussed a conference agenda with Claude, when you send Cowork to build the budget and logistics document, it already knows the headcount, city, and speakers. You do not have to brief it separately.

How the memory store is structured

The previous system condensed everything Claude had learned about you into a single summarised note. The new architecture replaces that with a set of short, structured files organised by topic. You can see the full list in Settings, under Memory, and you can edit or delete any individual item.

This granularity matters. If your company changed its name, you correct it once in the relevant file and every future conversation picks up the right name automatically. You are not hunting through a monolithic blob of text trying to find the line that needs changing.

Memory is also now created in real time rather than at the end of a conversation. If you mention mid-conversation that your project deadline has moved to September, Claude updates its memory then and there. You do not need to wait until the session ends or remind it next time.

You decide what gets remembered, including sensitive topics

By default, Claude does not store information related to health, race, ethnicity, religious beliefs, politics, gender identity, or similar areas. Anthropic is treating these as sensitive by default and keeping them out of memory unless you actively choose otherwise.

If you want Claude to remember that you are gluten intolerant when it is helping you with meal planning, or that you hold particular views relevant to the advice you are asking for, you can enable “include sensitive topics in memory” in Settings. With that setting on, each time Claude saves something in one of those categories, you will see a notice.

There is also a hard floor that cannot be unlocked by any setting. Certain information is never saved regardless of your preferences: Social Security numbers, government ID numbers, criminal history, immigration status, and anything that would violate Anthropic’s Acceptable Use Policy. Sensitive topics opt-in applies going forward only; it does not retroactively update anything Claude previously stored.

You can pause memory collection entirely or reset it at any time from Settings.

The legacy export deadline: 9 September 2026

Anthropic migrated users to the new system automatically, but acknowledges that some things may not have carried across cleanly. If you think Claude has forgotten something it used to know, go to Settings, then Memory. Until 9 September 2026, you will see an option to export your legacy memory. You can then paste that export back into Claude, highlight the part that seems to have been lost, and ask it to re-incorporate that information.

This is a limited window. If you relied heavily on Claude’s previous memory and want to make sure nothing was dropped in the migration, it is worth checking before that date.

Who gets this, and what stays separate

The unified memory is enabled by default for Free, Pro, and Max plan users on web, desktop, and iOS and Android (update to the latest app version to get it). Teams and Enterprise organisations have memory off by default; admins will receive separate information about their rollout timeline, and those users remain on the legacy experience in the interim.

Claude Code’s memory is not part of this change. Anthropic confirmed it is staying separate for now, and said it had nothing to share about whether Claude Code would join the unified system in the future.

Chat search, which lets you search through past conversations, remains a paid-plan feature and is separate from memory.

A word on transparency and privacy

Claude is designed to tell you when stored memory is influencing a response, rather than drawing on it silently. Memory is scoped to your account. It is not shared with other users, and Anthropic states it is not visible internally except for safety and abuse review purposes.

The bigger picture

Claude’s memory has been rolling out gradually. It reached Team and Enterprise plans in September 2025, then Pro and Max in October 2025, and became available to Free users in March 2026. The 25 August 2026 update is the most significant structural change in that progression: moving from a single-surface, end-of-conversation summary model to a shared, real-time, topic-structured store that spans Claude’s two main product surfaces.

The practical effect is that Claude can start to feel like one continuous assistant across the things you ask it to do and the tasks you delegate to it, rather than two products that happen to share a name.