Claude Now Remembers Your Business. Audit What It Holds
Anthropic merged Claude chat and Cowork memory into one store on 25/08/2026, so context now carries across both. It is on by default on personal plans, which makes the settings page worth ten minutes.
Claude has been building a file on your business, and from last week it uses that file everywhere.
What happened
On 25/08/2026 Anthropic merged the memory behind Claude chat and Claude Cowork into a single store. Anything Claude picks up in a chat is there when Cowork runs a task in the cloud, and anything that surfaces during a Cowork task carries back into chat. Explain your metrics definitions once in a conversation and the next document Cowork builds uses them without a rebrief.
Two other changes shipped alongside. Claude now writes memories during a conversation rather than summarising it at the end, so a detail mentioned this morning is already in place this afternoon. And everything held is visible as a list of short files under Topics in Memory settings, where each one can be read, edited or deleted. Correct your company name in one file and every conversation after that gets it right.
Sensitive subjects, including health, race, ethnicity, religious beliefs, politics and gender identity, are not stored unless you switch that on. The setting is not retroactive and can be turned off again. Some categories are never stored at all, whatever you choose: government identification numbers, criminal history, immigration status, and anything that breaches the Acceptable Use Policy.
Availability is the detail most write-ups skipped. Memory is on by default on Free, Pro and Max plans, across web, desktop and mobile. On Team and Enterprise it is the reverse: an admin has to make it available first, and each person still has to turn it on.
Why it matters for UK businesses
That split is worth pausing on. A sole trader or a five-person firm on individual Pro accounts gets the default-on behaviour with nobody supervising it. The organisations with the least governance capacity get the most automatic accumulation, and the ones with a compliance officer get the opt-in. If your firm sits in the first group, the setting is now your job.
There is a UK GDPR point too. Memory files that name people, a client contact, a supplier, a member of staff, are personal data you are holding. The visible, editable file list actually helps, because you can show what is stored and remove it on request. That is only comfortable if you have looked at it.
Then the practical limit: memory belongs to an account, not to a company. Three people using Claude in the same business build three separate pictures of it, and none of them can see the others.
How Adevious AI sees it
Turn it on, then write it rather than let it grow. Spend twenty minutes seeding memory deliberately in week one: registered company name, house style, date and currency conventions, how you name clients, how you define your numbers. A written memory is an asset you control. An accumulated one is a record you have never read.
Two cautions before anyone gets carried away. Because memory is per account, nothing propagates across a team, so this does not shorten onboarding for a new starter and it will not standardise output across your staff. Anyone hoping for shared institutional knowledge is looking at the wrong feature. And leave sensitive topics switched off on any account used for work. The convenience of Claude recalling a dietary requirement is small, and the record it creates on a business account is not something you want to explain later.
We would also treat the Topics list the way you treat a shared drive, with a monthly look rather than an annual panic. Real-time memory writing means the volume grows faster than the old end-of-conversation summaries did.
One thing to do this week
Open Settings, then Memory, then Topics. Read the list properly. Delete anything wrong, out of date, or belonging to a client rather than to you. Then write in three or four things you are tired of repeating, starting with how your business describes itself.
If you run a team, decide the position before someone asks you about it: on or off, individual accounts or organisation accounts, and who reviews what builds up.
Claude knowing your context is genuinely useful. Claude knowing it by accident is a different thing. Ten minutes in the settings page is the difference between the two.
If you would like help setting a sensible memory policy, or working out whether your firm belongs on individual or Team plans, get in touch with Adevious AI.
Sources: https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context https://techcrunch.com/2026/08/25/claude-cowork-finally-remembers-what-you-told-the-app-in-chat/ https://www.engadget.com/2243753/claude-memory-now-works-across-both-chats-and-cowork-sessions/ https://www.theregister.com/ai-and-ml/2026/08/25/claude-and-cowork-now-share-what-they-know-about-you/5292412 https://siliconangle.com/2026/08/25/anthropic-updates-claudes-memory-to-enhance-customization-and-protect-sensitive-topics/
Acronyms: AI: Artificial Intelligence GDPR: General Data Protection Regulation UK: United Kingdom