Anthropic’s Claude gets a major memory upgrade — transparent, compartmentalized & editable

Anthropic has rolled out a substantial upgrade to Claude’s memory features for paying users. The update adds more transparent, editable memory syntheses, per‑Project memory compartments, and the ability to import and export saved memories from other chatbots — all designed to make Claude more useful while giving users clearer control over what the assistant remembers.

The new memory set is optional and controlled via settings: Max subscribers can enable it today, with Pro access following in the coming days. Anthropic says Claude will learn a user’s preferences and work patterns over time so it can provide more helpful, context‑aware responses.

What’s new

• Transparent syntheses: Claude shows an “actual synthesis” of what it has recorded instead of vague summaries, so users see exactly what the model remembers.
• Editable memory via conversation: You can correct or remove saved items simply by telling Claude in chat.
• Per‑Project compartments: Projects create separate memory spaces, preventing personal and work contexts from bleeding into each other.
• Import & export: You can import saved memories from ChatGPT or Gemini and export Claude’s saved context to other AI platforms.

Safety & testing

Anthropic says it conducted extensive testing to avoid increased sycophancy or harmful conversation patterns from the new memory behaviors. The company refined memory functions based on those tests, making targeted adjustments to balance usefulness and safety.

Why it matters

Memory is a differentiator among chatbots — persistent context can boost productivity by avoiding repetition and tailoring suggestions. Anthropic’s emphasis on transparency, compartmentalization and conversational editing addresses common user concerns about control and privacy, while import/export features make Claude more interoperable with other assistants.

If you’re privacy‑minded, note that the feature is optional and must be explicitly enabled. Consider trying it in one Project first to see how Claude’s memory fits your workflow before expanding its use.

Read more: Anthropic · Engadget coverage.

Discussion: Would you enable an AI assistant’s persistent memory if it’s transparent and editable, or do privacy concerns keep you cautious?