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

Anthropic has opened an early research version of Claude Duet, a new app where its Claude AI works as an active assistant. The tool points out what needs attention, completes tasks, and keeps working even after users leave.

The launch quickly caught the attention of Binance founder Changpeng Zhao (CZ). “Now, it controls us,” he wrote on X, followed by a laughing emoji. However, Anthropic says Claude Duet still asks for approval before sending messages, making payments, or signing anything for a user.

What Can Claude Duet Do While You Are Away?

According to the company, Claude Duet runs on its own virtual computer with its own web browser. This means users do not need to keep their laptops or phones turned on.

Every morning, it sends users a short report based on connected apps such as Gmail, Slack, and QuickBooks. It then checks those apps every few hours and contacts users only when something needs their attention.

The agent can also handle larger tasks, such as scheduling meetings and creating dashboards, sales forecasts, and web apps.

Claude Duet uses Claude Opus 5.5 and brings over the saved memory, skills, and connected apps from a user's Claude account. This feature adds to the debate about the costs of switching AI tools while keeping their memory.

Who Has the Final Say When an AI Agent Acts?

Anthropic says the agent shows users what it plans to do and waits for their approval before sending a message, making a payment, signing a document, or deleting anything.

Some actions are not allowed even if users approve them. These include trading securities and moving money between accounts. The agent also never sees passwords directly, and a safety system blocks requests to share information such as card numbers.

However, the risks of AI agents acting on their own are real. Earlier this month, Anthropic revealed that, during a web-browsing test in July, an agent running Claude Haiku 4.5 sent Philadelphia police a false report of a murder.

CZ raised the same question about control in his response to the launch announcement.

Anthropic also warns that Claude Duet may misunderstand a conversation or make a mistake when deciding what needs urgent attention.

For now, people in the US who subscribe to Pro or Max can request access through a waitlist at no extra cost. Anthropic has not announced when the app will become available to more people. Team and Enterprise subscribers cannot use it yet.

The launch comes after Anthropic released its cheaper Haiku 5.5 model earlier this month. It also comes weeks before Anthropic's planned stock market listing in November. Investors have reportedly discussed a company value of nearly $1.5 trillion.

In the end, Claude Duet will help show whether people are ready to trust AI agents with their everyday work.