By Shekhar Chandran | Published August 26, 2026 | 7 min read | Facts verified against Anthropic’s official Claude release notes and Help Center memory documentation (support.claude.com), plus reporting from TechCrunch and SiliconANGLE.
📅 August 26, 2026 · 🕐 7 min read · 🗂️ AI Updates
Claude’s memory can now follow you out of a chat window and straight into a Cowork task. Anthropic shipped the fix on August 25, 2026. Until that day, Claude’s memory and Claude Cowork kept separate notebooks. You’d spend twenty minutes in chat explaining a project’s scope, hand the follow-up work to Cowork, and watch it ask the same questions from scratch. That gap is closed now, for cloud-run Cowork sessions anyway.
Here’s what this piece gets into: what changed, how the memory system works day to day, which plans have it switched on, where the privacy lines sit, how it stacks up against ChatGPT’s approach, and what it’s worth if you’re an Indian user stitching chat research together with Cowork execution.
This page covers: the August 25, 2026 memory-unification update, how continuous topic-based memory works, plan-by-plan availability, privacy/deletion controls, and a Claude-vs-ChatGPT memory comparison.
This page does not cover: Claude’s full INR pricing breakdown (see our Claude AI Pricing India 2026 guide) or a from-scratch Cowork setup walkthrough (see our Complete Claude Cowork Tutorial).
Jump to a section:
- What changed on August 25, 2026
- How Claude’s memory works now
- Which plans get it, and the catch
- Privacy: what’s excluded, what’s never stored
- Claude vs ChatGPT: two different memory bets
- 🇮🇳 What this means if you’re working from India
- My take
- AIInsider verdict
- Quick answers
- Related articles
What changed on August 25, 2026
Anthropic’s own release notes put it plainly for that date: “Memory now works across chat and Cowork in the cloud. Everything Claude remembers is listed under Topics…” The Help Center explains what that means in practice: “What Claude remembers from your chats is available when you hand it a task in Cowork in the cloud, and what comes up in a Cowork task carries back to chat.”
That’s the whole update, and it’s smaller than it sounds. But the friction it removes is real. Claude has had memory since September 11, 2025 (Team and Enterprise got it first, Pro and Max followed on October 23, 2025), and until this week it lived in two separate silos depending on which surface you were using. Research a client brief in chat on Monday, ask Cowork to draft the follow-up emails on Tuesday, and Cowork had no idea the conversation happened.
One memory now, instead of two.
How Claude’s memory works now
The mechanic underneath changed too, not just the sharing. Claude used to summarize a conversation for memory once it ended. Now, per Anthropic’s documentation, Claude saves memory as individual topics while you’re still chatting, not as one summary written after the fact. It’s a meaningfully different design. Memory updates continuously, so a task handed to Cowork mid-conversation already reflects what you just said instead of what you said an hour ago before the chat wrapped up.
| Old behavior (pre-Aug 25, 2026) | Current behavior |
|---|---|
| Memory summarized at end of chat | Memory saved topic-by-topic, live, during the chat |
| Chat memory and Cowork memory separate | Shared, but only for cloud-run Cowork |
| Each Claude Project keeps its own memory scope | Unchanged, still isolated per project |
| Manage memory in Settings | Unchanged, Settings → Memory, plus tell Claude directly in-chat |
You can review, edit, or delete any stored topic from Settings → Memory, or just tell Claude to “forget that” or “remember this instead” mid-conversation. Either way, the change takes effect starting with your next conversation.
Which plans get it, and the catch
| Plan | Memory on by default? | Where |
|---|---|---|
| Free | Yes | Web, desktop, mobile |
| Pro | Yes | Web, desktop, mobile |
| Max | Yes | Web, desktop, mobile |
| Team | No | Org owner has to switch it on |
| Enterprise | No | Org owner has to switch it on |
Here’s the catch. Memory syncing between chat and Cowork only works when Cowork is running in the cloud. Anthropic’s documentation says flatly that locally-run Cowork sessions don’t get memory at all. If you run Cowork on your own machine, a common choice if you’d rather not pay for cloud compute, this update changes nothing for you.
Searching past chats is a related but separate feature, and it’s still limited to Pro, Max, Team, and Enterprise. Free users don’t get it.
Privacy: what’s excluded, what’s never stored
Health concerns, race and ethnicity, gender identity, religion, political beliefs: Claude won’t store any of that by default. You have to explicitly turn it on, and Claude tells you when it saves something sensitive.
A separate, harder line exists underneath that setting. Government IDs, Social Security numbers, criminal history, immigration status, and anything that breaks Anthropic’s Acceptable Use Policy never get stored, full stop. There’s no toggle that overrides it.
Incognito chats sit outside memory entirely and don’t show up in chat search, which is handy for a one-off sensitive query you don’t want shaping future answers. One catch for Team and Enterprise users: incognito chats are still subject to your organization’s data retention policy and can turn up in admin data exports. Incognito isn’t the same as invisible to your employer.
Want out completely? Settings → Memory → Reset memory deletes every stored topic for good; there’s no undo. Pause is the gentler version: it keeps what’s already there and just stops adding anything new.
Claude vs ChatGPT: two different memory bets
We’ve covered how ChatGPT’s memory works in detail already, and it’s worth being specific about the contrast instead of just waving at “they’re similar.”
| Claude | ChatGPT | |
|---|---|---|
| Storage model | Continuous, topic-by-topic, saved live | Saved Memories (explicit) + Chat History Reference (RAG-based, implicit) |
| Transparency feature | View/edit/delete each topic in Settings | “Memory Sources” (May 2026) shows exactly which saved memory, chat, or file influenced an answer |
| Cross-product sync | Chat ↔ Cowork (cloud only), as of Aug 25, 2026 | No agentic-tool equivalent to sync with |
| Default-on for free tier | Yes | Varies by ChatGPT tier |
Honestly, they’re solving different problems. ChatGPT’s Memory Sources feature is about transparency: showing you why it said what it said. Claude’s August update is about fragmentation: making sure two products from the same company stop acting like strangers. They’re not really competing on the same axis, so “who has better memory” is a harder question to answer than most comparison posts let on.
🇮🇳 What this means if you’re working from India
🇮🇳 Freelancers and solo operators. If your workflow already looks like “research and draft strategy in Claude chat, then hand repetitive execution to Cowork” (the pattern in our Cowork automation guide), this removes an annoying re-briefing step every time you switch between the two. The saving isn’t dramatic on its own, but it’s the kind of small friction that adds up across a full day of client work.
🇮🇳 If you run Cowork locally to save on cost. This update doesn’t reach you. Weigh that trade-off with eyes open rather than assuming you already have the new behavior.
My take
Anthropic is calling this a memory upgrade, and technically it is. Functionally, it reads more like fixing a bug that never should have shipped. Two products from the same company, both marketed as able to “remember” you, and neither one told the other what it knew. That was always going to be the first complaint once Cowork adoption grew past the early testers. What I find more interesting than the fix itself is the design choice underneath it. Moving from end-of-chat summarization to live, continuous topic-saving is a bigger architectural shift than the coverage this week is giving it credit for, and it’s the part that’s likely to matter more a year from now.
Still, the cloud-only restriction is a real gap, not a footnote. Anthropic hasn’t said whether local Cowork will ever get memory access, and until it does, plenty of privacy-conscious and cost-conscious users get none of this.
AIInsider verdict
What we liked:
- Fixes a gap between chat and Cowork that users had been complaining about since Cowork launched
- Continuous topic-saving is a real architectural change, not just a marketing refresh
- Sensible default privacy posture: sensitive categories need opt-in, and a hard-blocked list of data is never stored no matter what
- On by default for Free, Pro, and Max, so most people don’t need to go hunting for a toggle
What holds it back:
- Cloud-only. Locally-run Cowork gets nothing from this update
- Team and Enterprise users don’t get it by default; someone has to remember to flip the org-wide switch
- No public timeline from Anthropic on closing the local-Cowork gap
Our recommendation: if you’re already paying for Pro or Max and use both chat and cloud-run Cowork, there’s nothing to set up. This is already working for you, and it’s worth two minutes in Settings → Memory to see what Claude has actually kept. If you run Cowork locally by choice, don’t expect this update to change anything for you yet.
Quick answers
Does Claude’s memory now work the same in Cowork and chat?
Yes, as of August 25, 2026, but only when Cowork runs in the cloud. Locally-run Cowork sessions get no memory access.
Which Claude plans have memory turned on by default?
Free, Pro, and Max have it on by default across web, desktop, and mobile. Team and Enterprise have access to it but it’s off by default until an org owner enables it.
Can I stop Claude from remembering sensitive topics?
Yes. Health, religion, political beliefs, race/ethnicity, and gender identity are excluded from memory by default. Claude only stores them if you explicitly opt in, and it notifies you when it does.
Can I delete everything Claude remembers about me?
Yes. Settings → Memory → Reset memory permanently deletes every stored topic, and there’s no getting it back.
Is incognito mode the same as having memory off?
Not quite. Incognito chats skip memory and chat search, but for Team/Enterprise accounts they’re still subject to your organization’s data retention policy and can show up in admin data exports.
Related articles
- Claude AI in India: The Complete 2026 Guide — the pillar guide this article belongs to: pricing, product lineup, and competitive positioning for Indian users.
- Claude AI Pricing India 2026: Official INR Rates Explained — exact Free/Pro/Max/Team rupee pricing, not repeated here.
- Complete Claude Cowork Tutorial (2026) — a full step-by-step walkthrough of setting up Cowork automations.
- What Is Claude Cowork? Complete Beginner’s Guide (2026) — start here if you haven’t used Cowork before.
- ChatGPT Memory Feature 2026 — How It Works, What’s New & Setup Guide — the comparison piece referenced above, in full.
Published August 26, 2026. Facts verified against Anthropic’s official Claude release notes and Help Center memory documentation (support.claude.com), with additional reporting cross-checked against TechCrunch and SiliconANGLE coverage published August 25, 2026. AIInsider.in is independent and not affiliated with Anthropic.