Claudes geteilte Erinnerung hat keine Mauer zwischen Chat und Cowork

Cryptopolitan · 2026-08-26T14:39:36.000Z

Anthropic hat am Dienstag die Erinnerung seines Claude-Chatbots und des Cowork-Assistenten zu einer gemeinsamen Erinnerung zusammengeführt. Dabei wurde das System so eingestellt, dass sensible Themen wie Gesundheit, Politik und Geschlechtsidentität weggelassen werden, es sei denn, die Nutzerin bzw. der Nutzer schaltet sie ein. Die Änderung gilt für alle, die bei Claude einen kostenlosen, Pro- oder Max-Tarif haben. Dabei ist die Erinnerung jetzt standardmäßig aktiviert und folgt der Nutzerin bzw. dem Nutzer von zwanglosem Chat bis zu automatisierten Arbeitsaufgaben. Eine Erinnerung, die zwischen Chat und Cowork umherwandert Cowork, der Assistent von Anthropic für Büroarbeiten wie das Umwandeln einer Tabelle in einen Bericht, und der konversationsbasierte Chatbot von Claude hielten die Erinnerungen der Nutzerin bzw. des Nutzers zuvor getrennt.

Anthropic on Tuesday unified the memory of its Claude chatbot and Cowork assistant into a shared memory. It set the system to leave out delicate topics like health, politics, and gender identity unless a user turns them on.
The change applies to anyone on Claude’s Free, Pro, or Max plans, where memory is now on by default and follows the user between casual chat and automated work tasks.
One memory that travels between chat and Cowork
Cowork, the assistant Anthropic built to do office work like turning a spreadsheet into a report, and Claude’s conversational chatbot kept separate memories of a user.
Anthropic said the two now share one memory. A detail that was mentioned in chat is available when Cowork later runs a task in the cloud, and anything that surfaced during a Cowork job flows back the other way.
The update removes the need to rebrief an AI on facts it should have. Anthropic’s own example has Cowork writing an update for a manager, and knowing who she is and how she likes updates written because the user told the chatbot earlier.
Anthropic’s blog says Claude will not save personal or sensitive material to memory. Health, race, ethnicity, religious beliefs, politics, and gender identity are on the company’s list, among similar areas.
A user who wants those remembered can enable a setting called “include sensitive topics in memory,” and then Claude will notify the user whenever it saves something in one of those categories.
Opting in only changes what Claude saves going forward. Nothing is saved retroactively, and the toggle can be turned off at any time.
Anthropic said Claude will not save Social Security numbers or government ID numbers, criminal history, immigration status, or anything that violates its Acceptable Use Policy and will tell the user when it refuses to save such information.
No split between the two, and Code stays out
Anthropic said there is no way to keep Chat and Cowork memories separate now that they run on one system, and that separate accounts are the only way to wall them off.
Short of that, the company pointed to fully pausing memory, chatting incognito, or deleting individual topics manually.
Claude Code, Anthropic’s software-development tool, keeps its own separate memory for now. The company said it had nothing to share on whether that will change.
Anthropic also changed the way memories form. Claude doesn’t summarize a conversation after the fact but adds topics as a chat is still happening, so a changed deadline discussed mid-conversation is available the next time around.
All that is retained is kept in a list of editable files under Topics in Memory settings.
The memory is switched on by default across web, desktop, and mobile for Free, Pro, and Max subscribers, although Cowork is not offered on Free accounts.
Admins decide if memory is available for Team and Enterprise users, and it remains off for individual users until they choose to turn it on, Anthropic said. Mobile users on iOS and Android need the latest app version to receive the update.
If you're reading this, you’re already ahead. Stay there with our newsletter.