Anthropic Opens a Waitlisted Research Preview of Claude Duet, a Proactive Agent That Must Ask Before It Sends, Pays, Signs or Deletes
Anthropic opened a waitlisted research preview of Claude Duet, a standalone app for US Pro and Max subscribers in which Claude Opus 5.5 works in the background on its own virtual machine, sends a morning brief and checks connected apps every few hours. Duet must get approval before sending messages, making payments, signing documents or deleting anything, and some actions are blocked even with approval. CNBC frames it as a challenge to Meta's Muse and OpenAI's Dots.
Anthropic's always-on agent is defined as much by what it may not do as by what it can. On Wednesday, October 21, Anthropic opened a research preview of Claude Duet, an app in which Claude works as a proactive agent: it decides what needs the user's attention, works through tasks in the background and, by Anthropic's design, checks before acting on the user's behalf. Access is by waitlist, starting with Pro and Max subscribers in the US, and Duet shares those plans' usage limits. CNBC says it will likely challenge OpenAI's Dots and Meta's Muse.
What Duet does
Anthropic says Duet is built around Claude Opus 5.5 and is aimed at professionals and small-business owners. It runs as a standalone iOS and Android app and on claude.ai, and carries over a user's Claude memory, skills and connected apps. Each morning it sends a short brief on the most important items across connected apps; it checks those accounts every few hours, which Anthropic says is to avoid using up plan limits faster; and it works on tasks on a virtual machine with its own browser, so the user's laptop or phone need not be on. Anthropic says it can use the more than 6,000 connectors and plugins in the Claude Marketplace, and can buy things for a business through Stripe's Link digital wallet, which pays with a virtual card, or through Shopify's Shop Pay.
The permission design
Before Duet sends a message, makes a payment, signs a document or deletes anything, Anthropic says, it shows exactly what it is about to do and waits for approval. "A few things are off-limits even with your approval, like trading securities or moving money between accounts." Anthropic says Duet's actions are checked by its auto-mode classifier, that its credential handling follows the design of Claude Managed Agents so it "never directly sees your passwords or sign-in tokens," and that it ships with a new classifier and safety prompt designed to reject requests to share the strictest tier of information, including credit card numbers and government IDs. Red teams and an external firm penetration-tested it, Anthropic says, and it promises more safety detail "in the coming days."
Anthropic says the preview "still might misread a thread or miscalibrate what's urgent," that personal errands such as booking tickets are weaker than work tasks, and that Duet cannot yet show the user its browser. OpenAI launched Dots, its persistent agents, at DevDay in September; Meta said the same evening as Duet's launch that more than 10 million people in the US and Canada have set up Muse agents, CNBC reports.
Limits
- Duet is a waitlisted research preview; its safety and reliability claims are Anthropic's and have not been independently tested.
- The comparison with Muse and Dots is CNBC's framing.