Anthropic just closed one of the more annoying gaps in Claude’s design. Until August 25, 2026, anything you told Claude in a normal chat stayed there. If you then handed a task to Cowork, Anthropic’s agent for multi-step work, you were starting from zero. Cowork didn’t know your manager’s name, your writing preferences, or the project you’d spent three conversations explaining the week before.
That’s no longer true. Claude now runs on one memory system across chat and Cowork, on web, desktop, and mobile. What you tell it in one place shows up when you use the other.
What actually changed
Anthropic merged the two memory systems that previously ran in parallel. Before this update, Cowork kept its own session memory, separate from whatever Claude had picked up in regular chat. Now there’s a single store both experiences pull from.
A few concrete effects:
- If you’ve discussed a project across several chats, Cowork starts a related task already knowing the details, instead of you re-briefing it from scratch.
- Claude adds information to memory while a conversation is happening, not only after it ends, so newer context is available sooner.
- The same memory follows you between the web app, the desktop app, mobile, and the Claude for Chrome side panel.
Anthropic’s own framing, from its August 25 announcement: memory is meant to work “like one continuous assistant rather than separate tools.”
Sensitive topics are handled separately
The update also changes what Claude is allowed to store without asking. Health conditions, race, ethnicity, gender identity, religion, and political beliefs are excluded from memory by default. Anthropic added a dedicated toggle for this category, so nothing in that bucket gets saved unless you turn it on yourself.
This matters more than it might seem at first glance. A shared memory system is only useful if people trust what’s going into it, and separating sensitive categories is Anthropic’s way of making that an opt-in decision rather than a default.
You can see, edit, or delete what’s stored
Claude now organizes memory by topic, and that list is browsable in Settings. You can open any topic, correct it, or remove it entirely. If Claude picked up something wrong, or something you’d rather it forget, you’re not stuck waiting for it to fade out on its own.
For anyone migrating from the older memory experience, Anthropic is giving a window to export the legacy version: that option is available under Settings > Memory through September 9, 2026.
Who has it, and who doesn’t by default
| Plan | Memory status |
| Free, Pro, Max | On by default, across web, desktop, and mobile |
| Team, Enterprise | Off by default; admins opt in |
If you’re on a paid individual plan, you don’t need to activate it. If you manage a Team or Enterprise workspace, memory has to be turned on at the account level before anyone sees it.
How to check or change your settings
- Open Claude and go to Settings > Memory.
- Review the topics Claude has saved. Edit or delete anything you don’t want kept.
- Decide whether to enable the sensitive-topics toggle. It’s off by default, and turning it on is a deliberate choice, not a default you’re opting out of.
- If you were using memory before this update, use the export option before September 9, 2026, if you want a copy of the old version.
- Turn memory off entirely from the same settings page if you’d rather Claude not retain anything between sessions.
Why this matters if you use Cowork for real work
Cowork is built for multi-step jobs, research, drafting, running tasks in the background, not quick back-and-forth chat. The old split meant people doing that kind of work had to keep two mental models of what Claude knew, and re-explain context every time they switched tools. That friction is exactly what this update targets. Anthropic is betting that a single memory layer makes it more practical to move from thinking something through in chat to actually having Cowork execute on it.
Whether that plays out depends on how well the memory retrieval scales as people’s saved topics grow, something that’s easier to judge in a few months than on launch day.
Frequently asked questions
When did Claude’s memory update roll out?
Anthropic announced the change on August 25, 2026.
Does Claude remember things from Cowork in my regular chats?
Yes. Memory is now shared in both directions between chat and Cowork.
Is memory on by default?
Yes, for Free, Pro, and Max plans, across web, desktop, and mobile. Team and Enterprise plans keep it off until an admin enables it.
Does Claude store sensitive personal information automatically?
No. Topics like health, race, ethnicity, gender identity, religion, and political beliefs are excluded unless you turn on a specific setting to allow them.
Can I see exactly what Claude remembers about me?
Yes. Settings > Memory shows saved information organized by topic, and you can edit or delete any of it.
What happens to my old memory data?
If you were using Claude’s memory before this update, you can export the legacy version from Settings > Memory until September 9, 2026.
