TLDRocket
Sign in

Anthropic updates Claude’s memory to enhance customization and protect sensitive topics

SiliconANGLE Kyt Dotson Covered by 3 sources

Anthropic changed Claude’s memory so users can see, edit, or delete what it keeps. It also skips sensitive topics by default, which is the part that actually matters.

Based on reporting by SiliconANGLE, Kyt Dotson — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Anthropic is giving Claude a much more transparent memory. Users can now see what the assistant remembers topic by topic, then edit or delete those memories instead of treating them like some hidden state tucked away inside the model.

The bigger shift is what Claude won’t remember on its own. Anthropic says the system now avoids storing sensitive subjects by default, including health concerns, race, ethnicity, gender identity, religion and political beliefs. If someone wants Claude to keep that kind of material, they have to turn a toggle on first.

That memory also follows the user across Claude chat and Cowork, Anthropic’s desktop agent for multistep work. So whether someone is chatting on mobile, using the web, sitting at a desktop or handing work to Cowork, the same memory is supposed to carry over for months of context. The company’s pitch is simple: stop forcing people to explain the same manager, report style or metric convention over and over again.

Anthropic says Claude adds memories as conversations happen rather than waiting for a summary at the end. Tell it a project deadline moved to September, and later chats should already know that. The idea is convenience, but the controls matter just as much. If a memory looks wrong, users can open the interface and change it quickly.

There are still hard limits. Even with sensitive memory turned on, Claude will not save highly private data like Social Security numbers, government ID numbers, criminal history or immigration status. Anthropic also says it won’t store anything that would break its Acceptable Use Policy, which covers safety, emotional harm, privacy and identity.

My take — AI-written commentary, not fact-checked reporting

This is the right kind of AI memory feature: user-visible, editable, and a little paranoid about sensitive data. Memory is only useful if people can see what the machine thinks it knows; otherwise it’s just a creepy notebook with branding. More companies should borrow the boring part here and skip the mystical nonsense.

Read more about this at: SiliconANGLE

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.