Claude Memory Just Got a Major Upgrade: Cross-Chat, Editable, and Privacy-First

You told Claude about your team's Q3 roadmap in a chat last Tuesday. Then you opened Claude Cowork to actually build the deck — and Claude had no idea what you'd just discussed. That gap closed on August 25, 2026, when Anthropic merged the memory systems behind Claude's chat interface and Claude Cowork, so anything Claude learns in one surface now carries into the other. Alongside the merge, Anthropic shipped something more consequential than the memory itself: a Topics view that lets you see, edit, or delete exactly what Claude remembers, plus a sensitive-topics setting that decides whether Claude is allowed to remember things like your health, religion, or politics at all.

This is the most transparent memory system any major AI assistant has shipped to date, and it changes how you should think about using Claude for anything spanning multiple sessions — research threads, ongoing projects, or CCA exam prep that stretches over weeks. Here's exactly what changed, how the controls work, and what it means for how you use Claude day to day.

What Changed on August 25, 2026

Before this update, Claude's memory was fragmented by product surface. Memory built up inside regular chat conversations, but Claude Cowork — Anthropic's collaborative workspace for multi-step, file-heavy work — ran on a separate memory track. Ask Claude in Cowork about something you'd explained in chat the day before, and you'd get a blank stare.

Anthropic's fix, announced in its Memory release notes, does three things at once:

Unifies memory across chat and Cowork. A preference, project detail, or working style Claude picks up in one surface is now available in the other, on web, desktop, and mobile.

Surfaces memory as editable Topics. Settings → Memory now lists everything Claude has retained, grouped by topic, with edit and delete controls for each item — not just a blanket "clear all memory" button.

Adds a sensitive-topics toggle, off by default, that governs whether Claude can remember personal or sensitive subject matter at all.

Memory is on by default for Free, Pro, and Max plans. It's off by default for Team and Enterprise organizations, where admins control whether it's enabled — a sensible default given that Team and Enterprise workspaces often mix multiple people's data and compliance requirements.

How the Sensitive-Topics Setting Actually Works

This is the part worth understanding carefully, especially if you use Claude for anything personal or professionally sensitive.

By default, Claude does not store information related to:

• Health and medical history
• Race and ethnicity
• Religious beliefs
• Political views
• Gender identity and sexual orientation

If you flip on Include sensitive topics in memory inside Settings → Memory, Claude begins retaining items in these categories — and critically, it tells you each time it does. You get a visible notification whenever something sensitive gets added, rather than silent background retention.

Even with the toggle on, some categories are permanently excluded. Anthropic has confirmed Claude will not store:

• Social Security numbers or other government ID numbers
• Criminal history
• Immigration status

One more detail that matters if you're deciding whether to turn the toggle on: it isn't retroactive. Enabling sensitive-topics memory only affects what Claude remembers going forward — anything discussed before you flipped the switch was never captured in that category and won't suddenly appear in your Topics list.

Where to Find and Control Your Memory

Every remembered item now lives in one place: Settings → Memory → Topics. From there you can:

• Browse by topic — Claude groups memories into categories (work context, preferences, ongoing projects, etc.) instead of dumping a raw log.
• Edit an item — correct something Claude got wrong or update it as facts change, without deleting the whole topic.
• Delete an item — remove a specific memory without wiping everything else Claude knows about you.
• Turn memory off entirely — a global switch if you'd rather Claude start fresh every conversation.

This matters because the previous mental model for AI memory — mostly Anthropic's, but also true of competitors — was "trust the black box or turn it off." Item-level editing means you can keep the useful memories (your writing style, your project's tech stack, your recurring meeting cadence) while pruning anything inaccurate or no longer relevant, without an all-or-nothing choice.

Practical Use Cases: Chat + Cowork Memory in Action

The cross-surface unification is the part that actually changes daily workflows. A few concrete examples of what now works that didn't before:

Multi-week projects. Start scoping a project in chat, then move to Cowork to have Claude execute file operations, spreadsheet work, or multi-step research — without re-explaining context, naming conventions, or constraints you already established. Recurring reports. If you've told Claude in chat that your weekly status report always needs a specific format (bullet points, no more than 300 words, metrics table first), that preference now follows you into Cowork when you ask it to actually assemble the report. Exam and study prep. If you're working through Claude Certified Architect (CCA) practice material across sessions, Claude can now recall which topics you've struggled with in past chats and adjust how it explains concepts in a later Cowork session — instead of treating every session as a cold start. Team handoffs (with caveats). Because Team and Enterprise memory is off by default, admins need to explicitly enable it if they want this cross-surface continuity for their organization — worth flagging to your workspace admin if you're on a business plan and want the same behavior.

What This Means If You're Building on Claude

For developers and power users, this update has a few downstream implications worth noting:

• Don't assume ephemeral context. If you're prompting Claude for anything you don't want remembered — a one-off sensitive question, a hypothetical scenario, throwaway brainstorming — be explicit, or check whether it landed in your Topics list afterward.
• Audit memory before high-stakes sessions. If you're using Claude for client work, legal review, or anything where prior context could bias output, a quick glance at Settings → Memory → Topics before you start is now a reasonable habit, the same way you'd clear browser autofill before filling out someone else's form.
• Sensitive-topics is a deliberate opt-in for a reason. If your work touches health, legal, or HR contexts, leave the toggle off unless you have a specific reason to want that retention — it's off by default precisely because those categories carry the highest risk if mishandled.

This also lands squarely in the "Claude ecosystem fundamentals" territory that shows up on the Claude Certified Architect exam — understanding how memory, context windows, and session continuity differ across Claude's surfaces is exactly the kind of platform-literacy question CCA tests.

How Claude's Memory Compares to ChatGPT and Gemini

Memory isn't a new idea in AI assistants — ChatGPT has had a version of persistent memory since 2024, and Gemini ties memory to your broader Google account activity. What makes Claude's August 2026 update notable is the combination of three things landing together, rather than any single capability being unprecedented:

The sensitive-topics design is the real differentiator. Most assistants either remember everything indiscriminately or give you a single global toggle. Claude's approach — categorizing memory by sensitivity, defaulting the riskiest categories to off, and notifying you in the moment something sensitive gets captured — treats memory as a privacy surface that needs graduated controls, not just an on/off switch.

Common Questions About Claude's New Memory System

Does turning on sensitive-topics memory retroactively capture old conversations?

No. Anthropic has confirmed the setting only applies going forward. Anything you discussed before enabling it was never stored in that category and won't retroactively appear in your Topics list.

Can I use Claude's memory feature on a Team or Enterprise plan?

Yes, but it isn't automatic. Memory is off by default for Team and Enterprise workspaces, and a workspace admin needs to explicitly enable it before it starts working across your organization's sessions.

Does memory work the same way through the Claude API?

No — this update is specific to the consumer-facing Claude chat interface and Claude Cowork. Memory behavior for API-based integrations is a separate mechanism that developers implement themselves, typically by managing conversation history and retrieval outside of Anthropic's consumer memory system.

What happens if Claude remembers something incorrectly?

Go to Settings → Memory → Topics, find the item, and edit or delete it directly. You don't need to clear your entire memory to fix one wrong detail.

Is memory visible to other people in a shared workspace?

Anthropic's Team and Enterprise controls are designed around admin-level enablement rather than shared memory pools between individual users, but if data isolation between teammates is a hard requirement for your organization, confirm the specifics with your workspace admin before enabling memory broadly.

Key Takeaways

• Anthropic unified Claude's memory across chat and Cowork on August 25, 2026 — no more re-explaining context when switching surfaces.
• A new Topics view under Settings → Memory lets you see, edit, or delete individual memories instead of an all-or-nothing reset.
• Sensitive topics (health, ethnicity, religion, politics, gender identity) are excluded by default; enabling them is opt-in, non-retroactive, and notified in real time.
• SSNs, criminal history, and immigration status are never stored, regardless of settings.
• Memory defaults to on for Free/Pro/Max and off for Team/Enterprise, where admins control activation.

Next Steps

Want to go deeper on how Claude's underlying architecture — memory, context, tool use, and agent workflows — actually works under the hood? That's the core of what the Claude Certified Architect (CCA) practice tests at AI for Anything cover, with study guides built for exactly this kind of platform update. Check your current Memory settings at Settings → Memory in Claude, and start with the free CCA sample quiz to see where your Claude platform knowledge stands.

Sources:

• Anthropic — Memory release notes
• SiliconANGLE — Anthropic updates Claude's memory to enhance customization and protect sensitive topics
• TechCrunch — Claude Cowork finally remembers what you told the app in chat
• Engadget — Claude's memory now works across both chats and Cowork sessions