Claude can now compartmentalize as part of a major memory upgrade (www.engadget.com)

Anthropic has rolled out a major upgrade to Claude’s long-term memory for paying users, adding editable, transparent, and compartmentalized memory spaces. Claude will now not only remember past conversations but present an “actual synthesis” of what it’s stored (rather than vague summaries), learn user preferences and work patterns over time, and let users edit memories through chat. A new Projects feature creates distinct memory buckets so work and personal contexts don’t bleed into each other. Anthropic also enables import of saved memories from ChatGPT and Gemini and export to other AI platforms, giving users cross-platform portability. The feature is opt-in and begins with Max subscribers, followed by Pro users in the coming days. For the AI/ML community this matters because it advances practical, persistent personalization while addressing safety and privacy concerns. The visible syntheses and per-project isolation improve transparency and boundary control—key when deploying assistants in mixed professional/personal workflows—while import/export supports interoperability and data portability. Anthropic reports extensive testing to reduce sycophancy and harmful outputs and made targeted adjustments to memory behavior, highlighting ongoing challenges in persistent-model alignment. Overall, the update demonstrates a maturing approach to long-term context in chat models: more useful personalization, clearer user-facing accountability, and explicit controls to limit unintended knowledge transfer.