← Accueil actu | Salle du marché

Anthropic gives Claude users topic-by-topic control over AI memory

2026-09-10 · Source:deep-siliconangle.com/feed/?paged=8

Anthropic gives Claude users topic-by-topic control over AI memory

Anthropic's latest Claude update lets users view, edit and delete stored memories topic by topic, while sensitive subjects such as health concerns are excluded from memory by default.

Artificial intelligence startup Anthropic PBC announced on Tuesday a rework of how memory functions in Claude, its flagship AI product, giving users direct oversight of what the assistant retains about them. With the change, users can see everything Claude remembers, presented topic by topic, and can edit or delete any of that stored information.

The update also alters what Claude holds onto in the first place. Sensitive subjects are not stored by default. That protection extends to topics mentioned by users during conversations, including health concerns, race and ethnicity, among other categories, so that material in those areas does not accumulate in the assistant's memory over time.

The redesign moves memory management out of the background and into the hands of the user. Because recollections are surfaced topic by topic rather than as one undifferentiated log, a person can excise a single category of remembered material while leaving the rest of the profile untouched. The edit and delete controls apply to any of the content Claude has retained.

The default exclusion of sensitive material means health concerns, race and ethnicity and similar subjects raised in conversation will not persist in Claude's memory. Anthropic frames the combination of granular visibility, editing tools and protective defaults as a way to enhance customization while protecting sensitive topics.

Claude is the company's flagship AI offering, and its memory shapes how much of a user's prior interactions carry into future conversations. With the announcement, decisions about what is remembered — and what is forgotten — sit explicitly with the user, from topic-level review through editing and deletion.

Mots-clés: Anthropic · Claude · AI Memory · Data Privacy


À lire aussi


萬安算交所 AIXX · AIXX · Bourse compute vérifiée · Accéder à la salle