Claude Duet: Your AI Assistant That Works in the Background
Claude Duet is Anthropic’s new AI app that works in the background across your apps but asks for your approval before taking important actions.
Illustration made on Lumeta with GPT Image 2.5Make one like it
Anthropic released a research preview of Claude Duet on Oct. 22, 2026, a standalone AI agent that works in the background on cloud computers and checks connected apps for you. It matters because Anthropic is joining the new race for proactive assistants, but with a strict approval model: Duet can prepare work on its own, yet it must stop and ask before sending messages, making payments, signing documents, or deleting anything.
What Claude Duet does
Claude Duet is built on Claude Opus 5.5 and runs on iOS, Android, and the web at claude.ai. Anthropic says it sends a morning summary from connected apps, then only reaches out when something needs attention. Users can assign tasks by text or voice in one ongoing thread, and Duet works on its own virtual machine in the cloud, so your phone or laptop does not need to stay on. Through more than 6,000 connectors and plugins in the Claude Marketplace, it can connect to services including Gmail, Slack, Stripe, Docusign, and Shopify for jobs like scheduling, unpaid invoice follow-up, billing prep, payroll prep, dashboards, and forecasts.
Illustration made on Lumeta with GPT Image 2.5Make one like it
How Anthropic says it stays in bounds
Anthropic says Duet can read, draft, and prepare on its own, but it pauses for approval before any sensitive step. That includes sending a message, making a payment, signing a document, or deleting something. Bank transfers and securities trading are blocked completely, even with approval. The company also says Duet does not directly see passwords or login tokens, uses prompt injection defenses from Anthropic's browser-use research, and screens messages for possible leaks of sensitive data such as card numbers or ID details. Health and personal finance information is only meant to pass through when the user explicitly allows it.
Illustration made on Lumeta with GPT Image 2.5Make one like it
Limits and who can try it now
Anthropic says the early version can still misread message threads, misjudge urgency, and miss updates that arrive between checks. It only checks connected accounts every few hours, because those background checks count against the same usage limits shared with other Claude products. Anthropic also says Duet is weaker at personal errands like booking tickets, since the product is aimed at work. For now, access is limited to Claude Pro and Max subscribers in the US through a waitlist. Team and Enterprise plans are not included yet. That puts Claude Duet directly into the same emerging category as Grok Bot, Meta's Muse, and OpenAI's Dots, but with a more tightly controlled rollout.
Is Claude Duet available to everyone right now?
No. Anthropic says the research preview is available through a waitlist for Claude Pro and Max subscribers in the US. Team and Enterprise plans are not included yet.
Can Claude Duet act without asking first?
It can read, draft, and prepare work on its own, but Anthropic says it must ask for approval before sending messages, making payments, signing documents, or deleting anything.
Can Claude Duet move money between accounts?
No. Anthropic says bank transfers and securities trading are blocked entirely, even if a user tries to approve them.
Does Claude Duet run on my computer all day?
No. Anthropic says Duet works on a cloud-based virtual machine, so your own laptop or phone does not need to stay on while it works in the background.
