Claude's Memory Now Spans Chat and Cloud Cowork - Editable in Topics, Sensitive Subjects Off by Default

Anthropic announced on August 25 that Claude's memory now works across chat and Cowork running in the cloud. Everything Claude remembers is listed under Topics in Settings > Memory, where it can be edited or deleted. Memory is on by default for Free, Pro and Max, and off for Team and Enterprise.

Claude Anthropic Cowork memory privacy enterprise-AI

On August 25, Anthropic announced that Claude’s memory now works across chat and Cowork running in the cloud¹. The company’s blog says that starting today, the memory you use in chat is the same one you use in Claude Cowork, and that Cowork now has memory for the first time¹. The same change is recorded in the help center release notes dated the same day².

The announcement also covers Topics, a list of everything Claude remembers, and a setting that governs how sensitive subjects such as health or beliefs are handled¹². Claude has had a memory for a while. What changed is where the boundaries sit — how much it retains, how far that travels, and how a user can inspect and correct it. If you use Claude at work, those boundaries matter before the convenience does.

What You Said in Chat Carries into Cloud Cowork

The release note reads: “Memory now works across chat and Cowork in the cloud”². The blog and help article are more specific: what Claude remembers from your chats is available when you hand it a task to Cowork in the cloud, and what comes up in a Cowork task carries back to chat¹³. The example given is asking Cowork to draft an update for your manager, where it already knows who that is and how they like updates written¹³. Another: brainstorm a conference agenda in chat, and when Cowork builds the budget and logistics doc it knows the headcount, the city and the speakers¹.

One limit must not be dropped here. This sharing applies only to Cowork running in the cloud. The help article states that memory is shared between chat and Claude Cowork when Cowork runs in the cloud, and that Cowork sessions running locally on your computer do not use memory³.

Cowork gained cloud execution on July 7, 2026, when sessions and files began saving to the account so work could continue on any device (Claude Cowork arrives on mobile and web). Memory itself changed on July 10, moving from a daily post-conversation summary to individual categorized entries that Claude reads and updates during conversations². This announcement is where those two lines meet.

What It Remembers Is Visible in a List

What Claude remembers is listed under Topics in Settings > Memory, where each item can be opened and read, changed with the edit icon, or deleted¹²³. The help article says that fixing one topic applies the change to every conversation from then on³. You can also update memory from within a chat by telling Claude what to remember, change or forget, and the update applies to your next conversation³.

The help article lists what gets remembered: your role, projects and professional context; the people and places in your work and life; communication preferences and working style; technical preferences and coding style; and details of ongoing work³.

The storage model is described too. Rather than summarizing a conversation after it ends, Claude saves individual topics as you chat — mention that a deadline moved and the next conversation already knows³. Each project has its own separate memory space and dedicated project summary, kept apart from other projects and from non-project chats³.

Giving an agent per-matter context that the user can inspect and change is the same shape Harvey II adopted for legal work. On the development side, GitHub Copilot for JetBrains introduced memory that persists across agent sessions. Having a memory is becoming less of a differentiator than whether the user can open it up and look.

Health and Beliefs Are Not Stored by Default

The handling of sensitive subjects is the part of this update that lands closest to practice.

The help article states that by default Claude does not store topics related to personal or sensitive subject matter — health, race, ethnicity, religious beliefs, politics, gender identity and similar areas³. Only if you would rather not keep re-explaining them do you turn on Include sensitive topics in memory in Settings > Memory³.

Behavior after turning it on is spelled out. Claude saves such topics going forward; nothing from before the setting was enabled is saved retroactively³. Each time it saves one, a notice appears above the message box so you can review it or change your settings³. And if you decline the notice or turn the setting off later, items already saved on those topics are removed³.

Some things are never stored even on request. The help article names government ID numbers, criminal history, financial account numbers and immigration status, and says Claude will tell you when it cannot save something for this reason³. The blog describes the same category as sensitive identification numbers (SSN, government ID numbers and the like), criminal history, immigration status, and anything violating the company’s Acceptable Use Policy — none of which is stored even with the sensitive topics setting on¹.

The Default Depends on Your Plan

For anyone deploying this in an organization, the first thing to check is that the default differs by plan.

Both the release notes and the help article state that memory is on by default for Free, Pro and Max, and off by default for Team and Enterprise organizations, where an owner can turn it on²³. Members of Enterprise plans can only enable it individually once an owner has enabled it for the organization³. The blog puts the Team and Enterprise case as admins controlling availability for their organization, with memory off for individual users until they turn it on¹. The blog also states explicitly that saving sensitive topics is off by default¹.

There are two ways to switch it off. Pause memory keeps existing memory but stops Claude from using it or making new memories; conversations held while paused are not added later if you unpause³. Reset memory permanently deletes everything including project memories, and cannot be undone³.

A few points about data handling are worth noting. Memory is retained in line with existing chat data retention policies, and memory data is included in data exports³. Enterprise data retention policies apply to all memory-related data, including incognito chats³. And when a conversation expires or is deleted, memory entries generated from it are not removed — though individual memories can be deleted at any time³. If you have been assuming that clearing chat history clears memory with it, that sentence corrects the assumption.

Separately, past chat search — a different feature from memory — is available on the paid Pro, Max, Team and Enterprise plans and is enabled by default once rolled out to an account³. For conversations you would rather not keep, incognito chats are saved neither to memory nor to chat history³.

Migrated Users Have an Export Deadline

The help article opens with a notice. Anthropic has introduced an improved memory experience and migrated users off the legacy one, and if you believe Claude forgot something in that migration, an option to export your legacy memory appears in Settings > Memory until September 9, 2026³. Once exported, you paste the content back to Claude and highlight the part that may have been lost³.

That deadline is about two weeks out. If you have been using Claude at work since before July and have noticed it failing to recall something you are sure you told it, there is room to check whether the export option is showing in your settings.