Anthropic has entered the race to build personal AI agents, joining rivals OpenAI and Meta in an intensifying race to develop digital assistants that can anticipate individual users’ needs and take action without waiting for a prompt.Known as Claude Duet, Anthropic said that its always-on personal AI agent is designed for professionals and small business owners. Unlike OpenAI’s Dots and Meta’s Muse, Claude Duet is less capable of executing personal tasks like booking reservations. It can connect to more than 6,000 external apps, including Canva and DocuSign.The distinguishing feature of Claude Duet, according to Anthropic, is its judgment. Instead of waiting for instructions, Claude Duet sends users a morning briefing drawn from information in their email inbox, calendar, and other apps. It can also check those accounts every few hours and reaches out to the user only when something needs their approval or attention.Powered by Claude Opus 5.5, the AI agent operates in a cloud-based virtual machine with its own browser. This means that users do not have to keep their laptop or phone on when Duet is running tasks in the background. Duet can be used to carry out simple administrative tasks such as scheduling meetings to more complex ones, such as building dashboards with dozens of tools, creating financial or sales forecasts, prototyping, and building web apps.The personal AI agent is currently rolled out for early testers as research preview. Subscribers to Anthropic’s Pro and Max plans in the United States can join a waitlist to use the agent, which will not cost them extra beyond their subscription fee. Claude Duet will also be available as a standalone app on iOS and Android, and on claude.ai.Claude Duet is Anthropic’s latest bet on AI agents that can manage people’s working lives, moving beyond chatbots that simply answer prompts. It also highlights an intensifying battle over what could be the next major frontier in consumer AI. Last month, OpenAI debuted Dots following the triumphant rollout of Meta’s Muse agent.Google’s Gemini Spark offering is also described as “your 24/7 personal AI agent”. Meanwhile, Apple has stepped up efforts with the launch of its redesigned Siri AI, which it described as “a profoundly more capable and personal assistant.” There are also agents from startups such as Instinct, which raised $1 billion in its latest funding round.Story continues below this adHowever, the adoption of personal AI agents could be held back by barriers to agentic commerce and growing concerns over security risks, with so-called ‘rogue’ or misaligned AI agents breaching company systems and meddling with government websites.Claude Duet can also make purchases through Link, Stripe’s digital wallet, or Shopify’s Shop Pay, though users must approve each purchase. This is interesting given that a bigger battle is brewing among online service providers and consumer AI agent vendors.Also Read | OpenAI’s Dots explained: How its ‘always on’ AI agents work, what they can doShortly after Meta’s rollout of Muse, e-commerce giant blocked the personal AI agent from accessing its shopping site, citing privacy and security risks. In previous earnings calls, Amazon CEO Andy Jassy has indicated that the company is in talks with other firms wanting to run agents for commerce.As part of its safety measures, Anthropic said Claude Duet will show users exactly what it plans to do before sending a message or making a payment. Additionally, the AI agent is not allowed to take actions like trading stocks or moving money between accounts even with the user’s approval. Other actions are approved by Anthropic’s auto mode classifier which is also used in Claude and Claude Code.Story continues below this adIn terms of handling credentials, Duet never directly sees a user’s passwords or sign-in tokens, similar to other Claude-managed agents. A new built-in classifier is meant to keep the agent from sharing credit card numbers or government ID numbers. Other sensitive information such as health or personal finances, are let through only with explicit user instructions to do so.Also Read | Meta’s Muse AI agent is taking off: What makes it different, why Amazon blocked itTo protect against direct or indirect prompt injection attacks, Anthropic said it has added specific defenses in Claude Duet aided by its research of agents’ browser-use. It has run adversarial tests for Claude Duet’s security infrastructure, along with red teaming efforts and a penetration test by an external firm.Despite these precautionary measures, Anthropic acknowledged that Claude Duet is still in early stages and might misread a thread or miscalibrate what is urgent.