Anthropic's Claude Duet Connects to Thousands of Apps for Autonomous Tasks With Approval Gates

Anthropic has launched a research preview of Claude Duet, an artificial intelligence agent that performs tasks in the background without requiring a user’s computer to remain active.

The risk

The primary security consideration for this product is the scope of access granted to an autonomous system. According to SiliconANGLE, Duet is built on Anthropic’s framework for trustworthy agents, which includes an automatic classifier that checks messages for sensitive information leaks. The agent uses a dedicated safety system prompt to guard against data exposure. While the system operates independently, it is designed to prevent unauthorized actions by pausing for human verification on critical steps.

Who is exposed

The preview is available only to Pro and Max subscribers in the United States, according to MacRumors. Users must join a waitlist to access the service. The product is available via the web, a standalone iOS app, and Android devices, as noted by The Next Web. Anthropic states the tool is built for professionals and small business owners who manage complex workflows, making these groups the initial target audience for the technology.

Cost and compliance

There is no additional cost for Claude Duet, as it is included in existing subscriptions, according to The Next Web. However, tasks performed by the agent count toward the user’s existing plan limits. Anthropic blocks certain high-risk activities entirely, including trading securities and transferring money between accounts. For purchases, users can utilize virtual cards through Stripe’s Link wallet or Shop Pay, adding a layer of financial separation.

What should you do?

Monitor the agent’s activity logs regularly to ensure it is interpreting task urgency correctly, as Anthropic admits it may misjudge priorities. Review the specific apps you have connected to the agent, verifying that each integration is necessary for your workflow. Be aware that the agent cannot display its browser activity, so you cannot visually verify its steps in real time.

The background in plain language

Proactive AI agents differ from traditional chatbots because they initiate actions rather than waiting for commands. These systems typically operate in isolated virtual environments to perform tasks like browsing or data entry. To manage risk, vendors often implement approval gates for sensitive operations, ensuring humans remain in the loop for critical decisions. Teams usually start with limited app connections to test reliability before expanding the agent’s permissions.

Background: AI agents

AI agents are software systems that perceive their environment, make decisions, and execute actions to achieve specific goals. Unlike standard models, they maintain internal state and interact with external tools, creating unique security risks around tool execution, memory retention, and autonomous decision loops that require specialized controls.

Read the full guide: AI Agents Explained: Answers to Security and Architecture Questions

What to do and how to stay safe: Anthropic

• Audit connected applications to remove any that are not essential for the agent’s core tasks.
• Monitor usage limits closely, as background checks and task completion consume subscription quotas.
• Wait for the vendor to provide updates on browser visibility features before relying on the agent for complex verification tasks.
• Review the agent’s morning briefs daily to catch any misjudged priorities or errors early.

Step-by-step guide: AI Governance Explained: Building Rules for Autonomous Systems

General security guidance from the Vector Tribune newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Can Claude Duet access my passwords?

No, according to MacRumors, the agent does not see passwords or logins when given access to accounts.

Does Claude Duet require a new subscription?

No, it is included in existing Pro and Max plans at no extra cost, though it uses existing usage limits.

Can the agent make financial transfers?

No, Anthropic blocks trading securities and transferring money between accounts altogether.