Anthropic Integrates AI Agent into Chrome

Anthropic launches Chrome extension for AI agent Claude.

AI startup Anthropic has launched a Chrome extension for its AI agent based on the Claude model.

Initially, the digital assistant named Claude for Chrome is available to a group of 1,000 Max plan subscribers. A waiting list is open for other interested users.

By adding the assistant to Chrome, users can interact with it in a sidebar. The agent retains the context of all activities within the web browser and can autonomously perform actions, completing certain tasks.

Browsers are becoming a new battleground for AI labs, as they are the primary gateway to the internet. Previously, Perplexity launched its own AI-based solution, Comet. The company also proposed acquiring Chrome for $34.5 billion.

In April, OpenAI expressed interest in a similar deal. The developer of ChatGPT is also preparing to introduce its AI browser with functionality akin to Comet.

The Risks of Autonomous AI Agents

In a blog post, Anthropic warned that the increasing number of AI agents with browser access poses new security risks.

Earlier, the Brave team stated that the Comet agent could be vulnerable to indirect prompt injection attacks, where hidden code on a site might deceive the virtual assistant into executing malicious instructions when processing a page. It is claimed that the vulnerability has been fixed.

Users of Claude for Chrome can restrict the agent’s access to certain sites in the settings. By default, the service cannot interact with platforms providing financial services, adult content, or pirated material.

Before performing “high-risk operations such as publishing, purchasing, or transmitting personal data,” the agent requests confirmation.

Back in August, Coinbase developers Kevia Leffew and Lincoln Murr stated that autonomous agents will become the largest active users of the network of the second-largest cryptocurrency by market capitalization.