Anthropic opened a research preview of Claude Duet on Oct. 211. The app turns Claude into an agent that checks your messages and calendar on its own.
Duet drafts replies, books meetings and flags unpaid invoices before you ask. It waits for your approval on the steps that matter most1.
Pro and Max subscribers in the United States can join a waitlist for early access. Anthropic built the preview around Claude Opus 5.510.
Who Gets the Morning Brief
Each morning, Duet sends a short brief. It names what needs you most across the apps you connect to it1.
Amin Ebrahimi owns Shahs of Kabob, a South Florida restaurant with about 110 employees. He now runs his invoices through Duet, store by store. "It brings it all together," he said. "I don't need to open up 100 tabs"1.
Duet reaches more than 6,000 connectors and plugins in the Claude Marketplace1. It can check Stripe and QuickBooks for unpaid bills, run payroll in Gusto, and prepare contracts in DocuSign.
A connector is Anthropic's name for a link between Duet and one outside app, like Gmail or Shopify. Each one has to be switched on before Duet can use it, and a user can switch any of them off again1.
Duet checks those connected accounts every few hours, instead of around the clock. Checking nonstop would burn through a plan's usage limits faster than a normal chat does1.
A few jobs stay off the table, even with approval. Trading securities and moving money between a user's own accounts both stay off-limits1.
Three Launches in Six Weeks
Duet is the third agent of its kind this fall. Meta shipped Muse on Sept. 85. OpenAI followed with Dots on Sept. 29, at its DevDay conference6. Anthropic closed the gap on Oct. 211.
Meta, OpenAI and Anthropic Launch Agents 43 Days Apart
Launch dates of three companies' always-on personal AI agents, September and October 2026
Each pin marks one company's launch date for its always-on agent. The gap between Meta's first move and Anthropic's answer spans 43 days.
Sources: Anthropic [1]; Meta Newsroom [5]; CBS News [6]. Chart by Kelly Dennis.
The numbers behind this chart
| Date | Event |
|---|---|
| Sept. 8, 2026 | Meta launches Muse |
| Sept. 29, 2026 | OpenAI launches Dots |
| Oct. 21, 2026 | Anthropic launches Duet |
Muse became the most downloaded app in Apple's App Store within weeks of its release6. Meta starts Muse at $0, with paid tiers for heavier use5. OpenAI priced Dots inside its Pro plan, which starts at $100 a month. Claude Pro costs $20 a month, and Duet rides along free, already included2.
OpenAI Charges 5 Times Anthropic's Price for Its Agent
Cheapest monthly plan that includes each company's always-on agent, October 2026
Each bar is the lowest monthly price that unlocks the company's agent. Meta's Muse starts free, with paid tiers for heavier use.
Sources: Anthropic [2]; Meta Newsroom [5]; CBS News [6]. Chart by Kelly Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| OpenAI's Dots | $100 |
| Anthropic's Duet | $20 |
| Meta's Muse | $0 |
Adam Crisafulli leads the investment advisory firm Vital Knowledge. He read Dots as aimed first at offices, ahead of households. "This seems very much geared toward enterprise/professional users, not everyday consumers," he wrote the day OpenAI launched it6.
Anthropic's Case for Trust
Anthropic frames Duet as a partner first and a tool second. "It comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf," the company said in its launch post1.
A dedicated classifier reviews every message Duet sends. Anthropic built it to reject requests carrying credit card numbers or government identification outright. Health and finance details pass only when a user gives Duet explicit permission to share them1.
Duet's handling of passwords follows the design Anthropic already uses for its Managed Agents product, keeping a login token outside the model's reach1. Red teams and an outside firm penetration tested the system before launch, and Anthropic now runs it under a public bug bounty.
Meta built a different answer to the same problem. Tarek Sheasha, a vice president at Meta Superintelligence Labs, described a second agent called Sentinel that sits apart from Muse at the system level. "Every interaction with the outside world runs through a Sentinel which the agent can't override," he wrote4.
Sentinel alone approves Muse's connector actions and its outbound network traffic. It swaps in real credentials only at that boundary, so the real credentials stay outside Muse's reach4. Anthropic's classifier approves the same kind of action from inside Duet's own stack, instead of a separate watchdog process.
Why Security Researchers Push Back
Hed Kovetz is the CEO of Silverfort, an identity-security company11. He wants a personal agent, and he still keeps his own accounts away from one. "I would be careful about giving full access to some of these commercial AI agents," he told CBS News the same day Anthropic launched Duet. "I am aware of the fact that their security and privacy are far from being ready"7.
Simon Willison, a software security researcher, named the pattern Kovetz is describing. He calls it the lethal trifecta: an agent that reads private data, meets content an attacker could plant, and can also send messages out. "If your agent combines these three features, an attacker can easily trick it into accessing your private data and sending it to that attacker," he wrote8.
Duet fits that shape by design. It reads a user's inbox, acts on what it finds there, and sends replies and payments on its own1. A hidden instruction buried in one email is the exact threat Willison's framework describes.
Readers on Hacker News argued the point within hours of the launch. The discussion thread drew 134 points and 127 comments9. Anthropic's own post on X collected 19,996 likes and 958 replies the same day3.
What to do
Read a company's safety page before approving an agent's first payment or signed document. Anthropic's own FAQ already warns that Duet can misjudge what counts as urgent. Check its Tasks tab daily during the research preview1.
Keep an agent away from any account outside its own connector list. Anthropic's credential protections cover only the access it manages directly, apart from a password typed into a browser by hand.
Compare prices before you join a waitlist. Duet rides along free on a $20 Claude Pro plan, where OpenAI's cheapest Dots access runs $100 a month26.
By the numbers
- 6,000 connectors and plugins link Claude Duet to outside apps through the Claude Marketplace1.
- Claude Pro costs $20 a month, and it includes Duet free during the research preview2.
- OpenAI's cheapest Pro plan runs $100 a month, the entry point for one Dot6.
- Meta's Muse launched 43 days before Anthropic answered with Duet51.
- Shahs of Kabob, the restaurant Amin Ebrahimi runs with Duet's help, employs about 110 people1.
- Hacker News readers gave Duet's launch thread 134 points within hours9.
- Anthropic's own X post announcing Duet drew 19,996 likes the same day3.
What to watch
Watch for Anthropic to publish its red-team results in detail, the way Meta laid out Sentinel's design in public. An outside audit of Duet's classifier, tested against a real prompt injection, would show whether the approval gate holds under attack. A release for Team and Enterprise customers would show whether Anthropic trusts this design at a bigger company's scale. That kind of leak reaches more than one person's inbox.
Sources
- Anthropic, "Introducing Claude Duet," Anthropic, Oct. 21, 2026, https://claude.com/resources/articles/claude-duet
- Anthropic, "Claude Duet," Anthropic, Oct. 21, 2026, https://claude.com/product/claude-duet
- Claude (@claudeai), "We're opening a research preview of Claude Duet," X, Oct. 21, 2026, https://x.com/claudeai/status/2112984664256241837
- Tarek Sheasha, "How We Built Safety Into Muse," Meta Superintelligence Labs, Sept. 8, 2026, https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse
- Meta, "Introducing Muse: The World's First Personal AI Agent Built for Everyone," Meta Newsroom, Sept. 8, 2026, https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
- Mary Cunningham, "Sam Altman unveils "dots," OpenAI's new AI personal agent," CBS News, Sept. 29, 2026, https://www.cbsnews.com/news/sam-altman-openai-dots-chatgpt-agents-safety/
- Megan Cerullo, "Should you trust personal AI agents with your money and data? Here's what experts say.," CBS News, Oct. 21, 2026, https://www.cbsnews.com/news/personal-ai-agents-trust-security-risks/
- Simon Willison, "The lethal trifecta for AI agents: private data, untrusted content, and external communication," Simon Willison's Weblog, June 16, 2025, https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/
- Hacker News, "Claude Duet," Hacker News, Oct. 21, 2026, https://news.ycombinator.com/item?id=50186037
- Marcus Mendes, "Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet," 9to5Mac, Oct. 21, 2026, https://9to5mac.com/2026/10/21/anthropic-takes-on-chatgpt-dots-and-meta-muse-with-claude-duet/
- Silverfort, "Our story," Silverfort, Oct. 21, 2026, https://www.silverfort.com/company/