Anthropic launches a Claude AI agent that lives in Chrome

Anthropic is launching a research on a browser-based AI agent driven by its clad AI models, agency Declaration CLode agent for Chrome on Tuesday, on anthropic’s Max plan, travels to a group of 1000 customers, which cost $ 100 to $ 200 per month. The company is also opening A Waitlist For other interested users.

By adding an extension to Chrome, select users can now chat with Claud in the Cedica window that maintains the context of everything that happens in their browser. Users may also allow Claud Agent to take action in their browser and to allow them to complete some work on their behalf.

The browser is quickly turning into the next battlefield for AI labs, whose aim is to use browser integration to provide a more unprotected connection between the AI ​​system and their users. The confusion has recently launched its own browser, CometIt features an AI agent that can offload tasks for users. OpenAI is reported to be close to its own launch AI-powered browserRumors have been rumors that have the same features with the comet. Mean Chrome In recent months.

AI-powered browsers are the development racing, especially the Google’s Looming AntitRast case, where the final decision is expected any day now. The federal judge in the case has suggested that he could force Google to sell his Chrome browser. Submitted to confusion a Offers $ 34.5 Billion inappropriately for ChromeAnd OpenAI CEO Sam Altman suggested that his organization would be Is willing to buy it As well.

In the blog post on Tuesday, anthropologists warned that the rise of AI agents with browser access creates new security risks. Last week, Brev’s security team said it was shown that the comet browser agent could be risky Indirect prompt-injection attackWhere the hidden code of a website can persuade the agent to enforce malicious instructions when it processes the page.

(Head of the confusion of confusion, Jesse Duer told TechCrunch in an email that the weakness has been raised boldly.)

Anthropic says that this research is hopeful as to capture and use fancy security risks, but the company has already introduced several defenses against immediate injection attacks. The company says its intervention has reduced the success rate of the immediate injection attack from 23.6% to 11.2%.

For example, anthropologists say that users can restrict the browser agent of the cloud to access specific sites in the app settings, and the company has default from access to the clode from access to websites that provide financial services, adult content and pirated content. The agency also said that the browser agent of the cloud “will seek the user’s permission before taking high-risky steps such as publishing, purchase or sharing personal data.”

This is not the first promotion of anthropic in AI models that can control your computer screen. In October 2024, the company launched an AI agent that could Control your PC – However, at that time the test revealed that the model was very slow and incredible.

The powers of agent AI models have been quite improved since then. TechCranch has shown that modern browser-users are AI agents such as comet and Chatzipt agentIt is fairly reliable in offloading simple tasks for users. However, many of these agents still fight against more complex problems.