Anthropic opens Claude Duet
On 21 October 2026 Anthropic opened a research preview of Claude Duet, a separate app in which Claude works as a proactive agent: it sends a short brief each morning, watches connected apps, works in the background on a virtual machine of its own and asks for approval before it sends a message, pays, signs or deletes. Access is by waitlist, starting with US Pro and Max subscribers; the app runs on iOS, Android and claude.ai.
Why it matters
The page describes an agent that comes to the user first instead of waiting for a prompt, and builds the limits into the product: every send, payment, signature or deletion waits for approval, and trading securities or moving money between accounts is refused even with approval. Anthropic says it is designed around Claude Opus 5.5 and checks connected accounts only every few hours to spare usage limits. A reader cannot try it yet: it is a waitlist preview, and Anthropic says it can misread a thread or miscalibrate what is urgent.
What the page says. Claude Duet is aimed at professionals and small business owners. It carries over the user's memory, skills and connected apps from their Claude account, takes messages by typing or voice in one ongoing thread, and keeps working when the user steps away, on a virtual machine with its own browser. It uses connectors and plugins from the Claude Marketplace (the page says over 6,000); examples include Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva and Shopify. For purchases it can use Stripe's Link, which asks the user to approve the spend and pays with a virtual card, and Shop Pay; signing in with logins saved in 1Password is marked "coming soon". Limits the page names. Before it sends a message, makes a payment, signs a document or deletes anything, the agent shows exactly what it is about to do and waits for approval; trading securities and moving money between accounts are off-limits even with approval. It checks connected accounts every few hours, because constant checking would use up plan limits faster. Anthropic says personal errands such as booking tickets are weaker, that it cannot yet show the user its browser, and that it lives in its own app, separate from the main Claude app. Safeguards, in Anthropic's words. Actions are approved by its auto mode classifier; credentials are handled as in Claude Managed Agents, so the agent never sees passwords or sign-in tokens; a new dedicated classifier and safety system prompt check each message for sensitive information (the strictest tier, including card numbers and government identification, is refused; other sensitive information, such as health or finances, goes only when the user has said so explicitly); and red teams and an external penetration-testing firm tested the app. A separate article on safety, privacy and security is dated 23 October. Access. Included in Pro and Max plans with the same usage limits as the rest of Claude; the background checks count against them. The page says the preview starts "today" with Pro and Max users in the US and tells others to join the waitlist. What the record does not claim. Everything above is Anthropic's own description; no independent test of the product was read. The testers' quotations on the page were not used.