A.I. news · October 22, 2026

Claude Duet A.I. Checks Before It Sends, Pays or Deletes

Anthropic's new Claude Duet is an A.I. agent that works in the background, but it shows you what it's about to do before it sends, pays, signs or deletes.

What Claude Duet is and who can get it

✅ Anthropic has opened a research preview of Claude Duet. It's an app where Claude works as a proactive agent. ITmedia reports the release came on October 21, US time.^[1][2]

Anthropic says Duet brings you what needs your attention and completes tasks you ask for. It keeps working when you step away, and it checks with you before it acts.^[1]

Anthropic says it built Duet for professionals and small business owners.^[1]

It's a standalone app on iOS and Android, and it's also on claude.ai. It's included in Pro and Max plans with the same usage limits as the rest of Claude. Access requires joining a waitlist, and Anthropic is starting with Pro and Max users in the US.^[1]

What it does all day

🔎 Anthropic says Duet is designed around Claude Opus 5.5. It's built to decide what needs you, what can wait, and when to check with you first.^[1]

Each morning, it sends a short brief on the most important items across your connected apps. During the day, it keeps tabs on those apps and messages you only when something needs attention.^[1]

You can hand it jobs too, from scheduling meetings to building dashboards, forecasts, prototypes and web apps. It can use over 6,000 connectors and plugins in the Claude Marketplace.^[1]

It can check Intuit QuickBooks or Stripe for unpaid invoices and draft reminders. It can also buy things for your business using Link. Duet confirms the item, and Link asks you to approve the spend.^[1]

Duet runs on a virtual machine with its own browser, so your laptop or phone doesn't need to be on. It checks your connected accounts every few hours, because constant checking would use up your limits faster.^[1]

The safety guardrails

⚠️ Anthropic says Duet never directly sees your passwords or sign-in tokens. Before it sends, pays, signs or deletes, it shows you what it will do and waits for your approval.^[1]

Some things are off-limits even with approval: trading securities and moving money between accounts.^[1]

A new classifier is designed to reject requests to share credit card numbers and government identification. Health or personal finance details go out only when you explicitly say so.^[1]

Anthropic also says red teams and an external firm penetration tested Duet. It says Duet draws on its trustworthy agents framework, which also guides Claude Code and Claude in Chrome.^[1]

Our take: approval screens only help if you actually read them. A guardrail you click past is just a speed bump.

What early users and outlets say

Amin Ebrahimi owns Shahs of Kabob. He used Duet for accounting, support, marketing and invoices.^[1]

VentureBeast.Tech says OpenAI built Dots as a character and Meta built Muse around consumer errands. It also writes that Anthropic appears to have reused developer infrastructure for Duet, and reports that Claude Loop has a new name: Claude Duet.^[3]

The outlet also reports that Duet says its morning brief and check-ins run as Claude Code Routines. That's the agent describing itself, so we'd treat it as a clue, not a confirmed design.^[3]

VentureBeast.Tech notes Duet isn't watching your inbox second by second. Our take: the contest among always-on helpers could shift from who answers best to who you trust to act for you.^[3]

How to try it wisely

💡 If you're on Pro or Max in the US, join the waitlist. Anthropic will email you when Duet is ready.^[1]

VentureBeast.Tech advises connecting the minimum, starting with one inbox and one calendar. Remember that connecting your inbox means Duet reads it. You can disconnect any app it has access to, and you can tell it what you want to hear about and how often.^[3][1]

Our take: set your alert preferences on day one, or it may ping you too much or too little. Then try one boring chore, like chasing unpaid invoices, before bigger jobs.

Anthropic says background checks count toward your usage limits. Bigger tasks need more model turns and background work. It also says personal errands are weaker.^[1]

What's still unknown

Anthropic says Duet might misread a thread or miscalibrate urgency. It also can't yet show you its browser. Anthropic says it will share more safety details in the coming days.^[1]

Our take: one mistaken email or missed deadline could cost trust that's hard to win back. Watch how access, usage limits and the promised safety details play out.

💡 Our take: what this could mean for you

• Start with one inbox and one calendar, then add more once it earns your trust.
• Read every draft and charge before you approve it. Approval only protects you if you look.
• Test it on one boring chore first, and keep an eye on your plan's usage limits.

Sources

1. Introducing Claude Duet — Anthropic
2. Anthropic版常時稼働エージェント「Claude Duet」発表　Opus 5.5ベースで設計　Pro／Maxプランで提供 — ITmedia
3. Claude Loop Becomes Claude Duet: Anthropic's Agent Goes Mobile — VentureBeast.Tech