On Tuesday, Anthropic announced the launch of a research preview for a revolutionary browser-based AI agent powered by its Claude AI models. This innovative tool, named Claude for Chrome, is being rolled out to a select group of 1,000 subscribers enrolled in Anthropic’s Max plan, which ranges in price from $100 to $200 monthly. Additionally, the company has opened a waitlist for other interested users looking to access this cutting-edge AI technology.
By adding an extension to Chrome, chosen users can now engage in conversations with Claude through a sidecar window, which maintains context with everything occurring in their browser. This allows for a more interactive and user-friendly experience. Users can even grant the Claude agent permission to execute actions in their browser, enabling it to complete various tasks on their behalf.
The emergence of Claude for Chrome highlights the growing competition among AI laboratories to integrate AI capabilities into web browsers. Recently, Perplexity launched its own browser, Comet, featuring an AI agent designed to assist users by offloading tasks. Reports indicate that OpenAI is also nearing the release of its own AI-powered browser, rumored to have functionalities similar to Comet. Furthermore, Google has introduced Gemini integrations with Chrome in recent months, intensifying the race for advanced browser technologies.
The urgency surrounding the development of AI-powered browsers is compounded by Google’s ongoing antitrust case, with a final ruling expected shortly. A federal judge has hinted at the possibility of forcing Google to divest its Chrome browser. In a surprising move, Perplexity made an unsolicited $34.5 billion offer for Chrome, while OpenAI CEO Sam Altman has expressed the willingness of his company to make a similar acquisition.
In a blog post released on Tuesday, Anthropic underscored the potential safety risks associated with the rise of AI agents that have browser access. Recently, Brave’s security team identified vulnerabilities in Comet’s browser agent, specifically concerning indirect prompt-injection attacks. This type of attack occurs when hidden code on a website deceives the agent into executing harmful instructions during page processing. However, Jesse Dwyer, Perplexity’s head of communications, assured that the identified vulnerability has since been rectified.
Anthropic aims to leverage this research preview as an opportunity to identify and mitigate emerging safety risks. The company has already implemented several defenses against prompt injection attacks, successfully reducing the success rate from 23.6% to 11.2%. For instance, users have the option to restrict Claude’s browser agent from accessing specific websites within the app’s settings. By default, Anthropic has blocked Claude from visiting sites related to financial services, adult content, and pirated materials. Additionally, Claude’s browser agent is programmed to seek user permission before undertaking any high-risk actions, such as publishing, purchasing, or sharing personal data.
This isn’t Anthropic’s first venture into creating AI models capable of controlling computer screens. In October 2024, the company introduced an AI agent with the capability to operate PCs. However, testing at that time revealed that the model was often slow and unreliable, prompting further refinements in their technology.
As the competition heats up in the realm of AI-powered browsers, Anthropic's Claude for Chrome is set to play a pivotal role in shaping the future of web interactions, all while addressing critical safety concerns.
