CZ Reacts to Anthropic’s New Claude Duet AI Agent: ‘Now, It Controls Us’

Anthropic has opened a research preview of Claude Duet, a new app in which its Claude AI acts as a proactive agent. The tool flags what needs attention, finishes tasks, and keeps working after users step away.

The launch quickly drew a reaction from Binance founder Changpeng Zhao (CZ). “Now, it controls us,” he wrote on X, followed by a laughing emoji. However, Claude Duet still asks for approval before it sends, pays, or signs anything on a user’s behalf, Anthropic says.

What Can Claude Duet Do While You Are Away?

Claude Duet runs on its own virtual machine with its own browser, the company says. As a result, a user’s laptop or phone does not need to stay on.

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf. pic.twitter.com/kt8gAbWuX8

— Claude (@claudeai) October 21, 2026

Each morning, it sends a short brief from connected apps such as Gmail, Slack, and QuickBooks. After that, it checks those apps every few hours and messages users only when something needs them.

The agent also takes on bigger jobs, from scheduling meetings to building dashboards, sales forecasts, and web apps.

Claude Duet uses Claude Opus 5.5 and carries over memory, skills, and connected apps from a user’s Claude account. That transfer feeds into the debate over AI memory switching costs.

Who Has the Final Say When an AI Agent Acts?

Before the agent sends a message, makes a payment, signs a document, or deletes anything, it shows the planned step and waits, Anthropic says.

Some actions stay off-limits even with approval, including trading securities and moving money between accounts. In addition, the agent never directly sees passwords, and a classifier blocks requests to share data such as card numbers.

Still, the risks of AI agents acting alone are not theoretical. Earlier this month, Anthropic disclosed that a Claude Haiku 4.5 agent sent Philadelphia police a fake murder tip during a web-browsing test in July.

CZ touched on the same question of control in his reply to the launch post.

Now, it controls us. 🤣 https://t.co/32vHaGyRSN

— CZ 🔶 BNB (@cz_binance) October 21, 2026

Anthropic itself warns that Claude Duet may still misread a thread or misjudge what is urgent.

For now, access runs through a waitlist for Pro and Max subscribers in the US, at no extra cost. Anthropic has not said when Claude Duet will open more widely, and Team and Enterprise plans are not included yet.

The launch follows a cheaper Haiku 5.5 model released earlier this month. It also arrives weeks before Anthropic’s planned November listing, where investors have reportedly discussed a valuation near $1.5 trillion.

Ultimately, Claude Duet will show whether users are ready to trust AI agents with their everyday work.