Claude: Tell it once. Carry the context.

Anthropic says the memory used in Claude chat is now the same one used in Cowork. That makes the release less about a model remembering more and more about whether a piece of working context can survive the move from conversation to action.

The catch is obvious. Shared memory can sound like an opaque profile following us around. Anthropic’s answer is a Topics layer we can inspect, edit, delete, pause or reset.

When we move from a chat about a project into a Cowork task, the frustrating part is often the re-brief: the deadline, the tone, the people involved, the decision already made. Anthropic says those details can now travel through the same memory.

That can make a working session feel less fragmented, but it does not turn memory into a guarantee. The useful habit is still to check the context that matters before we rely on it.

Availability is part of the release, not the whole story. Anthropic says memory is on by default for Free, Pro and Max across web, desktop and mobile; on iOS and Android, we may need the latest app version for the newest updates.

For Team and Enterprise, Anthropic says administrators control availability and memory remains off until each person enables it. That makes checking the actual setting more useful than assuming the feature has arrived everywhere in the same way.

Anthropic says memory is stored as a list of Topics. That matters because a wrong detail should not become a permanent piece of working context: we can find the Topic, correct it or remove it.

Anthropic also says we can explicitly ask Claude to remember something specific. The control is most valuable when it is used as part of the work, rather than treated as a one-time privacy setting.

Everything Claude remembers is in a list of files under Topics in Memory settings, where you can read, edit, or delete each one.

Anthropic, “Claude’s memory works everywhere, and you decide what’s in it”

The shared-memory claim is deliberately narrow: Anthropic says chat and Cowork use the same memory. It should not be read as a promise that context flows across every Claude or Anthropic product.

Nor does continuity settle the accuracy question. When a detail carries weight, we still need to check it, especially before using it in a consequential task.

Anthropic says Topics update as we chat rather than only after a conversation ends. A project deadline can become part of the saved context; later, we can inspect the Topic, correct it, delete it, pause memory or reset it.

What happens inside the feature beyond those published controls is not specified here. The practical point is simpler: continuity is only useful when the record behind it remains reviewable.

In practice, we might explain a project change in chat, ask Cowork to act on it later, then open Topics to make sure the context is still right. If it is not, the useful action is correction or deletion, not faith that the system will resolve it on its own.

Sensitive-topic handling adds another reason to inspect settings. Anthropic says saving sensitive Topics is off by default, applies going forward if enabled, and does not include every category.

The value of shared memory is not that it carries everything forward. It is that the context we choose to carry remains visible enough to challenge.

Creator Broadcast synthesis from Anthropic’s published memory announcement

Anthropic’s announcement makes a practical case for continuity between chat and Cowork. The stronger reading is also the more cautious one: a saved detail should become easier to revisit, not easier to forget is there.

Before we lean on a remembered context, we can audit the Topics behind it. That keeps the feature grounded in an observable check rather than a vague promise of frictionless work.