Claude Memory Explained: What It Remembers and How to Control It

Claude has been keeping notes on you for a while. As of today it keeps better ones, in more places, and it finally shows you the file. On August 25, 2026, Anthropic rebuilt Claude's memory: it now saves what it learns as you talk instead of summarising conversations afterwards, it carries between the Claude chat app and Claude Cowork, and every single thing it has stored is a short file you can open, edit, or delete. It is free, it is already switched on, and the one setting most people will want to look at is the one about sensitive topics.

Key Takeaways

• It is free and already on. Anthropic says memory is "on by default on Free, Pro and Max plans across web, desktop, and mobile." There is nothing to buy and nothing to enable.
• It saves as you go. Claude "adds topics to memory as you chat, instead of summarizing conversations after they end." Say a deadline moved, and the next conversation knows.
• Health, politics and beliefs are excluded by default. Anthropic does not store "health, race, ethnicity, religious beliefs, politics, gender identity, and other similar areas" unless you deliberately turn them on.
• Some things are never stored, even if you ask. Anthropic's help centre names government ID numbers, criminal history, financial account numbers and immigration status. Claude tells you when it refuses.
• You can read every word of it. Settings, then Memory, then Topics. Each entry is a short file you can edit or delete, and a fix applies to every conversation afterwards.
• There is a deadline if Claude forgot something. Anthropic migrated users off the old memory system. Your legacy memory can be exported until September 9, 2026, and not after.
• Team and Enterprise are the opposite. Memory is off there until an owner turns it on for the whole organisation.

Do You Get It? Almost Certainly Yes

If you use Claude on a personal plan, you have this now, at no cost. This is one of the rare AI releases with no paid tier, no waitlist and no country list attached to it. The only people who have to do anything are those on a work account, where the decision belongs to an administrator rather than to you.

Two vendor pages describe Team and Enterprise slightly differently and it is worth knowing why. Claude's pricing matrix marks Memory as included on Team and Enterprise, while Anthropic's help centre says it is "off by default and can be turned on by an owner." Both are true: the pricing page is telling you the feature is in the plan, the help centre is telling you it is not running yet. Follow the help centre if you are wondering why you cannot see it.

Platforms are simple: web, Claude Desktop and Claude Mobile on iOS and Android. Anthropic notes you need the latest mobile app version to get the current behaviour.

What Claude Actually Saves

Claude saves the everyday working context that would otherwise need re-explaining, and it stores it as a set of separate topic files rather than one giant profile. Anthropic's own list is your role and professional context, the people and places in your work and life, your communication preferences and working style, technical and coding preferences, and details of ongoing projects.

The change worth understanding is when it saves. The old system summarised your conversations after the fact and refreshed a synthesis every 24 hours. The new one writes to memory during the conversation. Anthropic's example is the practical one: "Mention that your project deadline moved to September, and your next conversation already knows without you having to say 'remember this.'" You can still say it explicitly if you want something pinned.

Projects stay separate. Anthropic says each project "has its own separate memory space and dedicated project summary," so what Claude learns inside one project does not leak into your other chats.

The Sensitive Topics Switch

This is the setting to look at, and Anthropic has defaulted it in the cautious direction. By default Claude does not save anything it considers personal or sensitive: health, race, ethnicity, religious beliefs, politics, gender identity, "and other similar areas." If you never touch the setting, none of that is retained.

You can turn it on, and there is a real reason to. Anthropic's example is a food allergy: if you are tired of re-explaining a gluten allergy every time you ask for recipes, "include sensitive topics in memory" makes that stick. The switch lives in Settings, then Memory, and Claude also offers it via a one-time notice in chat the first time it declines to save something.

There is a hard floor underneath all of this. Anthropic's help centre says some information "is never saved to memory, even if you ask," and names government ID numbers, criminal history, financial account numbers, and immigration status. Claude tells you when it cannot save something for that reason. Anthropic's launch post frames the same list slightly differently, adding anything that breaches its usage policy; the help centre's version is the more specific one and was updated the day of the launch.

One Memory Across Chat and Cowork, With a Catch

Memory is now shared between the Claude chat app and Claude Cowork, so context you build up in conversation is there when you hand Cowork a task, and what Cowork learns comes back the other way. Anthropic's example: ask Cowork to draft an update for your manager and "it already knows who that is and how she likes updates written."

The catch is not in the launch post, but it is in the help centre, and it matters if you use Cowork locally. Memory is shared only when Cowork runs in the cloud. Anthropic is explicit: "Cowork sessions that run locally on your computer don't use memory." So if your Cowork work happens on your own machine, the shared-memory headline does not apply to you.

How to See, Fix, and Delete What Claude Knows

Everything is in one place, and it is readable. Go to Settings, then Memory, and look under Topics. Each entry is a short file. Select one to read it, use the edit icon to change it, or select Delete to remove it. Because the files are short and specific, correcting one is cheap and permanent -- Anthropic's example is fixing your company's old name once so every later conversation gets it right.

You do not have to go to settings at all if you would rather not. Tell Claude in the chat what you want it to remember, change, or forget, and Anthropic says the update applies to your next conversation.

Pause and Reset are genuinely different and the difference is easy to miss. Pause keeps everything Claude already has but stops it using or adding to memory, and conversations held while paused are not backfilled if you unpause later. Reset permanently deletes all memories including project memories, and Anthropic says plainly that "this cannot be undone."

One limit to be clear about: deleting a conversation does not delete what Claude learned from it. Anthropic says that when a conversation expires or is deleted, "related memory entries generated from it won't be removed" -- you have to delete those entries yourself in Settings.

The September 9 Deadline

If Claude seems to have forgotten something it used to know, this is why, and there is a time limit on fixing it. Anthropic has migrated people off the older memory system onto this one, and some context did not survive the move. Its help centre gives a way out: "Navigate to Settings > Memory and until September 9, 2026, you will see the option to export your legacy memory."

The recovery is manual but it works. Export the legacy memory, paste it back into Claude, and highlight the part that went missing so Claude can pick it up again. After September 9 that export option is gone.

There is also a quick way to tell which system you are on, which is useful because Anthropic's own help pages document both. If your settings show a Memory section, you are on the new system and everything above applies. If memory instead sits under Settings, then Capabilities, you are still on the legacy one -- Anthropic describes the rollout as gradual, so not everyone has moved yet.

Should You Care?

Leave it on if you use Claude for anything ongoing. Memory is the difference between an assistant that knows your job and one you brief from scratch every morning, and this version is the first where you can actually audit what it has concluded about you.

Check the sensitive topics setting if you talk to Claude about your health, your beliefs or your politics. It is off by default, which is the right default, but "off" is only reassuring once you have seen for yourself that it is off.

Turn it off if you would rather Claude stayed forgetful -- and use Pause rather than Reset unless you are certain, because Reset takes your project memories with it. For one-off private questions, incognito is the lighter tool and it is on every plan including Free.

Go and look at Topics either way. It takes two minutes and it is the only way to find out whether Claude has been carrying a wrong assumption about you into every conversation.

The honest verdict: this is a good release with an unusually clean consumer answer -- free, on by default, fully visible, and cautious about the categories that matter. The one thing with a clock on it is the legacy export, so if Claude has gone blank on something it used to know, deal with that before September 9. If you are still deciding which assistant to commit to, our guides to the best free AI chatbot and which AI chatbot you should pay for cover the parts that do cost money, and Claude Reflect is the monthly recap that this same memory setting switches on and off.

The bottom line: Claude's memory is free, already running, and now readable line by line in Settings, then Memory. Sensitive subjects like health and politics stay out unless you opt in, government ID numbers and financial account numbers stay out regardless, and if the migration lost something you have until September 9, 2026 to export it back.