Until this week, Claude had a small but genuinely annoying split personality. Whatever you taught it about yourself in a regular Claude Chat conversation stayed in Chat. Whatever you taught it while using Cowork - Anthropic's agentic tool for multi-step tasks like organizing files or running a research job in the cloud - stayed in Cowork. Jump from one to the other and Claude greeted you like a stranger, even though you'd explained your project, your preferences, and your deadline an hour earlier. On August 25, 2026, Anthropic closed that gap. Claude's memory is now unified across Chat and Cowork, and it updates live as you talk instead of waiting for a conversation to end. Here's what actually changed, what Anthropic is and isn't willing to remember about you, and the one real limitation this doesn't solve.
The announcement in brief: Anthropic's own blog post confirms Claude's memory store is no longer split between Chat and Cowork - anything Claude learns in one now surfaces in the other. Memory also now updates continuously during a conversation rather than only being generated afterward; Anthropic's example is a project deadline mentioned mid-chat showing up in your very next conversation, unprompted. It's on by default for Free, Pro, and Max plans across web, desktop, and mobile (update your app for the mobile version), though Cowork itself isn't available on Free. Team and Enterprise customers have it off by default until an admin turns it on. Sensitive categories - health, race/ethnicity, religion, political views, gender identity - are excluded from memory by default, and a fixed list (government ID numbers, SSNs, criminal history, immigration status) is never stored no matter what you toggle. Anthropic told The Register that the only way to fully keep Chat and Cowork memories separate is to run two different accounts.
What Actually Changed
The headline change is structural: one memory store instead of two. Before this update, Claude Chat and Claude Cowork behaved like two apps that happened to share a login. Tell Chat you're building a React dashboard for a client and prefer terse code comments, and that context simply wasn't there when you later handed a file-organization job to Cowork. Anthropic's fix is exactly what it sounds like - Chat and Cowork now read from and write to the same memory, so context flows in both directions between the two experiences.
The second change is about timing, and it's arguably the more useful one day-to-day. Memory used to be generated in batches after a conversation wrapped up. Now it updates continuously, mid-conversation. Anthropic's own example is a good illustration of why this matters: mention that your project deadline moved to September, and you don't have to say "remember this" - your very next conversation already reflects the change. For anyone who uses Claude across a scattered workday - a quick Chat question at 9am, a Cowork task at noon, back to Chat in the evening - that live-update behavior is what actually makes the unification feel seamless rather than theoretical.
There's a notable exception worth flagging if you also write code with Anthropic's tools: Claude Code, the standalone coding-agent product, is not part of this merge. It keeps its own separate, independent memory system for now - so if your workflow spans Chat, Cowork, and Claude Code, the third one still won't know what the first two do.
Who Gets It, and Who Has to Turn It On
Rollout is broad but not universal by default. Memory is on out of the box for Free, Pro, and Max plans, on web, desktop, and mobile - though mobile users need to be on the latest app version, and it's worth remembering Cowork itself isn't part of the Free plan in the first place, so Free users get the unified-memory benefit only within Chat until they upgrade. Cowork sessions run remotely in the cloud, which is also why your sessions and files follow your account across desktop, web, and mobile rather than being tied to one device.
Team and Enterprise accounts work the opposite way: shared memory is off by default, and it stays off until an administrator explicitly enables it for the organization. That's a sensible default for workplace deployments - a company doesn't want an employee's personal Cowork tasks quietly informing what Claude assumes about them in a shared Chat workspace unless IT has decided that's acceptable.
The Privacy Trade-Off, and the Controls You Actually Get
Merging two memory stores into one naturally raises the stakes on what that single store contains, and Anthropic's blog post spends real space on this. There are two separate lines of defense. The first is a default exclusion list you can override: health data, race and ethnicity, religious beliefs, political views, and gender identity are not stored in memory unless you deliberately opt in through a toggle in Settings > Memory. The second is a hard exclusion list with no toggle at all - government-issued ID numbers, Social Security numbers, criminal history, and immigration status are never stored in memory, regardless of your settings or what you tell Claude to remember.
Beyond those two lists, Anthropic gave memory its own dedicated settings page. Settings > Memory now shows what Claude remembers organized topic by topic, and you can view, edit, or delete any individual topic rather than being stuck with an all-or-nothing wipe. Incognito mode remains available for any conversation you don't want touching memory at all - those chats aren't saved to memory and don't show up in your conversation history afterward.
The honest caveat, and it's Anthropic's own admission rather than outside criticism: when The Register asked how to keep Chat and Cowork memories fully separate - say, because you use Cowork for work and Chat for personal questions and don't want the two mixing - Anthropic's answer was that the only complete solution is a second account. Pausing memory, going incognito, or deleting specific topics are the partial workarounds available inside one account, but none of them recreates the old hard wall between Chat and Cowork. If keeping those two contexts genuinely separate matters to you, plan around that limitation rather than assuming a settings toggle covers it.
The Real Limitation: It's Claude-Only
The most practical catch for anyone who doesn't live inside a single AI vendor: this memory unification is entirely internal to Anthropic. It fixes the split between Claude Chat and Claude Cowork, but it does nothing for the far more common split most people actually deal with - the one between Claude and every other model they use. If you route coding questions to Claude, run a quick fact-check through Gemini, and occasionally open GPT-5.6 for a second opinion, none of what Claude now remembers about your project or preferences travels with you to those other tools. Switch vendors and you're back to re-explaining your context from scratch, exactly as before - the unification only closed the gap within Anthropic's own products, not across the wider set of models people actually route work through day to day.
- Claude's memory is now unified across Chat and Cowork - information learned in one now surfaces in the other, instead of being stored separately.
- Memory updates continuously during a conversation now, not just after it ends - a mentioned deadline change shows up in your very next chat.
- On by default for Free, Pro, and Max across web, desktop, and mobile (update your app); off by default for Team/Enterprise until an admin enables it.
- Health data, race/ethnicity, religion, political views, and gender identity are excluded from memory by default - opt in via a Settings > Memory toggle if you want them included.
- Government ID numbers, SSNs, criminal history, and immigration status are never stored in memory, with no setting able to override that.
- A new Settings > Memory page lets you view, edit, or delete what Claude remembers topic by topic; incognito chats never touch memory or appear in history.
- Anthropic told The Register that a second account is the only way to fully keep Chat and Cowork memories separate - pausing memory or deleting topics are partial fixes, not a full wall.
- Claude Code keeps its own separate memory system and isn't part of this merge.
- This memory lives entirely inside Anthropic's products - it doesn't follow you into GPT-5.6, Gemini, Grok, or any other model.
01Does this mean Claude Chat and Cowork are now the same product?
No - they're still separate tools with different purposes (Chat for conversation, Cowork for agentic multi-step tasks run in the cloud). What changed is that they now draw on and contribute to one shared memory store instead of two separate ones, so context carries between them.
02Is unified memory on by default for everyone?
For Free, Pro, and Max plans, yes - across web, desktop, and mobile (mobile requires the latest app update), though Cowork itself isn't available on Free. For Team and Enterprise accounts, it's off by default and requires an administrator to turn it on.
03Can I stop Claude from remembering sensitive things about me?
Health data, race/ethnicity, religion, political views, and gender identity are already excluded by default - you'd have to actively opt in to include them. Separately, government ID numbers, SSNs, criminal history, and immigration status are never stored no matter what you choose. You can also review, edit, or delete anything Claude has remembered from Settings > Memory, or use incognito mode for a conversation you don't want touching memory at all.
04Can I keep my Chat and Cowork memories completely separate if I want to?
Not fully within one account. Anthropic told The Register that using two separate accounts is the only way to guarantee full separation. Inside a single account, pausing memory, using incognito mode, or deleting specific memory topics are the available partial workarounds.
05Does this unified memory carry over if I also use ChatGPT, Gemini, or Grok?
No. This memory system is specific to Anthropic's Claude products. It doesn't sync with or transfer to other AI vendors, so switching to GPT-5.6, Gemini, or Grok for a task means starting that model's own context from zero, regardless of what Claude already knows about you.
For anyone whose actual workflow doesn't stop at Claude - flipping between it and GPT-5.6 for a second opinion, or Gemini for a quick search-grounded answer, or DeepSeek to keep costs down on a big job - this update is a reminder of where the walls still are: solved within Anthropic, unsolved between vendors. That's the gap LumiChats is built around, putting Claude alongside GPT, Gemini, DeepSeek and other leading models under one login with pay-per-day pricing under $1/day, so switching models for the right task doesn't also mean starting your context over from nothing.
