Discussion

Claude memory is now editable, and a new analysis says the hard questions are still open

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#2767

Claude's memory is no longer a rolling diary you simply have to trust. Since this summer's updates, remembered items are individual, categorised and editable as Topics, and a new analysis argues that progress is real while the enterprise questions remain open.

Anthropic Watch analysis

What happened

Archie Ang of MemTensor published the piece on Hugging Face today, working from Anthropic's release notes. The changes landed in two steps. On 10 July, Claude's memory stopped being a daily summary and became individual categorised entries that the model reads and updates in conversation. On 25 August, memory began working across Chat and Cowork in the cloud, remembered items became editable or deletable as Topics, and health and beliefs were excluded unless a user explicitly enables sensitive topics.

The defaults carry the organisational story. The release notes, as the article reads them, put memory on by default for Free, Pro and Max plans and off by default for Team and Enterprise organisations. Consumers get memory automatically; companies have to choose it.

– Categorised entries, not a daily summary Since 10 July, Claude reads and updates individual memory entries in conversation rather than one rolling digest.

Why it matters

– Whether editable, categorised memory with sensitive-topic gates spreads to rival assistants and becomes table stakes.

Our read

– How Team and Enterprise administrators configure memory now that it is off by default on those plans.

What to watch

  • Claude now lets you edit what it remembers about you. Would agent memory earn your trust once it is editable inside the app, or only when you can export the whole record and take it elsewhere?
  • What actually changes for users if the reported move ships.
  • Which named facts later reporting confirms or walks back.

Discussion spark: Claude now lets you edit what it remembers about you. Would agent memory earn your trust once it is editable inside the app, or only when you can export the whole record and take it elsewhere?

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.