anthropic launches a claude ai agent that lives in chrome

Anthropic Launches Claude for Chrome: A New AI Agent

Anthropic announced on Tuesday the initiation of a research preview for an AI agent accessible through a web browser. This agent is powered by the company’s Claude AI models.

Known as Claude for Chrome, the agent is currently being deployed to a select group of 1,000 subscribers who are enrolled in Anthropic’s Max plan. This plan is priced between $100 and $200 monthly.

A waitlist has also been established to accommodate further user interest.

Browser Integration and Agent Capabilities

By installing a Chrome extension, chosen users are now able to engage in conversations with Claude within a side panel. This panel intelligently retains context related to all browser activity.

Furthermore, users can grant the Claude agent the authority to perform actions directly within their browser, thereby automating certain tasks.

The Growing AI Browser Landscape

The web browser is rapidly evolving into a key area of competition for AI developers. These companies are striving to create more integrated and seamless experiences between AI systems and end-users.

Perplexity recently introduced its own browser, Comet, featuring an AI agent designed to handle tasks on behalf of the user.

Reports indicate that OpenAI is nearing the release of its own AI-enhanced browser, anticipated to offer functionalities similar to those found in Comet.

In parallel, Google has been integrating Gemini features into Chrome over recent months.

Antitrust Concerns and Browser Acquisitions

The development of AI-powered browsers is gaining urgency, particularly in light of Google’s ongoing antitrust case. A final ruling is expected soon.

The presiding judge has indicated a potential requirement for Google to divest its Chrome browser.

Perplexity made an unsolicited offer of $34.5 billion for Chrome, and OpenAI CEO Sam Altman expressed his company’s willingness to consider an acquisition.

Safety Considerations and Prompt Injection Risks

In a blog post released on Tuesday, Anthropic highlighted the emerging safety challenges associated with AI agents possessing browser access.

Last week, Brave’s security team identified a potential vulnerability in Comet’s browser agent. This vulnerability could allow for indirect prompt-injection attacks, where malicious code embedded on a website could manipulate the agent.

Jesse Dwyer, Perplexity’s head of communications, confirmed to TechCrunch that this vulnerability has since been addressed.

Anthropic’s Safety Measures

Anthropic intends to utilize this research preview to proactively identify and mitigate novel safety risks. The company has already implemented several defenses against prompt injection attacks.

These interventions have demonstrably reduced the success rate of such attacks, from 23.6% to 11.2%.

Users have the ability to restrict Claude’s browser agent from accessing specific websites through the application’s settings.

By default, Claude is blocked from accessing sites containing financial services, adult content, and pirated materials.

Before undertaking potentially risky actions – such as publishing content, making purchases, or sharing personal information – Claude’s browser agent will request explicit user permission.

Previous AI Agent Development

This is not Anthropic’s initial venture into AI models capable of controlling a computer’s interface.

In October 2024, the company launched an AI agent designed to control a PC. However, initial testing revealed limitations in speed and reliability.

Improvements in Agentic AI

Significant advancements have been made in the capabilities of agentic AI models since that time.

Recent evaluations by TechCrunch indicate that contemporary browser-based AI agents, including Comet and ChatGPT Agent, are generally reliable in handling straightforward tasks.

However, these systems often encounter difficulties when confronted with more intricate challenges.