Anthropic launches a Claude AI agent that lives in Chrome

The company announced on Tuesday that Antarbur launches a research inspection of the artificial intelligence agent on the browser -backed Claude Ai. The agent, Claude for Chrome, is offered to a group of 1000 subscribers in the MAX plan for the month of Anthropor, which ranges between $ 100 and $ 200 per month. The company also opens a waiting list for other interested users.

By adding an extension to the Chrome, users can now chat with CLADE in the Sidecar window that maintains the context of everything that happens in their browser. Users can also give the Claude agent’s permission to take action in their browser and complete some tasks on their behalf.

Soon the browser became the following battlefield for AI Labs, which aims to use the browser integration to provide more smooth communications between artificial intelligence systems and its users. Perplexity recently launched its own browser, Comet, which features an AI factor that can empty the tasks of users. Openai is said to be about to launch his own browser, which is rumored to contain similar features for the guilt. Meanwhile, Google has launched Gemini with Chrome in recent months.

The race to develop browsers with artificial intelligence browsers in particular, given the control of the monopoly that waves on the horizon of Google, in which a final decision is expected on any day now. The federal judge in the case suggested that he may force Google to sell Chrome. Al -Hujir made an unwanted offer of $ 34.5 billion of chrome, and Openaii CEO Sam Altman suggested that his company be ready to buy it as well.

In the Tuesday’s blog post, Antarubor warned that the rise of artificial intelligence agents with access to the browser is the risk of new safety. Last week, the Prave’s Security team said that it was found that the Comet Browser may be vulnerable to indirect immediate injection attacks, as the hidden symbol on the web site can be deceived in implementing harmful instructions when processing the page.

(Jesse Duaire, the head of communications at Perplexity, told Techcrunch in an email that the courage collected had been fixed.)

Man says she hopes to use this research inspection as an opportunity to arrest and address new safety risks, however, the company has already provided many defenses against immediate injection attacks. The company says its interventions reduced the success rate of immediate injection attacks from 23.6 % to 11.2 %.

For example, Anthrop says that users can reduce the Claude browser agent from accessing certain sites in the application settings, and the company has prevented, by default, Claude from accessing web sites that provide financial services, adult content and pimented content. The company also says that the Claude browser will ask the user’s permission before “taking high -risk measures such as publishing personal data, buying or sharing personal data.”

This is not the first anthropologist in artificial intelligence models that can control the computer screen. In October 2024, the company launched Amnesty International’s agent who can control your computer – however, the test revealed at the time that the model was very slow and unreliable.

The capabilities of artificial intelligence models have improved somewhat agent since then. Techcrunch has found that artificial intelligence agents who use a modern browser, such as Catgpt Agent, are somewhat trusted in emptying the simple tasks of users. However, many of these systems are still struggling with more complex problems.