Claude Duet safety design: own VM, confirmation before key actions
Claude Duet runs on Opus 5.5 in its own VM with a browser, never sees your passwords or tokens, and asks before sending, paying, signing or deleting; US Pro and Max users can join the waitlist.
Anthropic launched Claude Duet yesterday, a proactive agent that comes to you instead of waiting for a prompt. It is built around Claude Opus 5.5 and works in its own virtual machine with a browser, so it keeps going when you step away. Each morning it sends a short brief from your connected apps, then messages you only when something needs your attention. It never directly sees your passwords or sign in tokens, an auto mode classifier approves its actions, and it asks before it sends a message, makes a payment, signs a document or deletes a file. A new classifier also refuses to share card numbers or government IDs. It is a research preview for US Pro and Max subscribers, by waitlist, on iOS, Android and http://claude.ai. This puts Anthropic up against OpenAI's Dots and Meta's Muse.
This post quotes another post: x.com/claudeai/status/2112984664256241837