Anthropic launches a Claude AI agent that lives in Chrome

Anthropic launches a research review of a browser-based AI agent powered by his Claude AI models, the company announced On Tuesday. The agent, Claude for Chrome, rolls out to a group of 1,000 subscribers on the maximum plan of Anthropic, which costs between $ 100 and $ 200 a month. The company also opens one waiting list For other interested users.

By adding an extension to Chrome, users can now select with Claude Chat in a Sidecar window that retains the context of everything that happens in their browser. Users can also give the Claude Agent permission to take actions in their browser and complete a few tasks on their behalf.

The browser quickly becomes the next battlefield for AI Labs, which are aimed at using browser integrations to offer more seamless connections between AI systems and their users. PerTlexity recently launched his own browser, Comet, with an AI agent who can discharge tasks for users. OpenAi is reportedly close to launching its own AI-driven browser, which is said to have comparable functions as a comet. In the meantime, Google has launched Gemini integrations with Chrome in recent months.

The race to develop AI-driven browsers is especially urgent in view of the threatening antitrust case of Google, in which a definitive decision is expected every day. The federal court in the case has suggested that he can force Google to sell his Chrome browser. PerTlexity submitted an unsolicited range of $ 34.5 billion for Chrome, and OpenAi CEO Sam Altman suggested that his company would also be willing to buy it.

In Tuesday’s blog post, Anthropic warned that the rise of AI agents with browser access results in new safety risks. Last week the Brave security team said it discovered that the browser agent could be vulnerable Indirect fast injection attacksWhere hidden code on a website can mislead the agent to perform malignant instructions when it processed the page.

(The head of the communication of Pertlexity, Jesse Dwyer, told WAN in an e -mail that the vulnerability that is good has been raised has been resolved.)

Anthropic says it hopes to use this research preview as an opportunity to catch and tackle new security risks; However, the company has already introduced several defenses against fast injection attacks. The company says that its interventions lowered the success rate of rapid injection attacks from 23.6% to 11.2%.

For example, Anthropic says that users can limit Claude’s browser agent to gain access to certain sites in the app settings, and the company has blocked Claude as standard to access websites that offer financial services, content for adults and illegal content. The company also says that Claude’s browser agent will request permission from users before “take actions with a high risk, such as publishing, buying or sharing personal information.”

This is not the first trip from Anthropic in AI models that can control your computer screen. In October 2024, the company launched an AI agent who could control your PC – the testing at that time revealed that the model was fairly slow and unreliable.

The possibilities of Agentic AI models have since improved quite a bit. WAN has found that the modern browser-usual AI agents, such as Comet and Chatgpt agent, are fairly reliable when unloading simple tasks for users. However, many of these agent systems are still struggling with more complex problems.