Anthropic launches Claude AI for Chrome
Anthropic is rolling out a new experiment in AI-assisted browsing, announcing Tuesday that it has launched a research preview of a Chrome extension powered by its Claude AI models. Dubbed Claude for Chrome, the agent is initially being made available to 1,000 Anthropic Max subscribers—users paying between $100 and $200 per month—with a waitlist now open for broader access.
The extension allows users to chat with Claude in a sidecar window that maintains context of what’s happening inside their browser. More notably, users can grant the agent permission to take certain actions on their behalf, effectively turning Claude into a lightweight browser co-pilot.
The Browser Becomes the Next AI Battleground
The move underscores how the web browser has become the latest competitive frontier for AI labs. Perplexity recently unveiled Comet, its own AI-powered browser, while OpenAI is rumored to be preparing a similar product. Google has also introduced Gemini integrations into Chrome in recent months, further tightening the race.
That competition comes as the browser market itself faces potential upheaval. A federal judge is expected to soon issue a ruling in the Justice Department’s antitrust case against Google. The judge has suggested that Google may be forced to divest Chrome—a possibility that has drawn interest from rivals. Perplexity made an unsolicited $34.5 billion offer for the browser, while OpenAI CEO Sam Altman has indicated he would be open to buying it as well.
Balancing Capability and Risk
Anthropic acknowledged in its blog post that AI agents with browser access pose new risks. Last week, Brave’s security team warned that Comet’s browser agent was susceptible to indirect prompt-injection attacks, in which hidden instructions on a website could manipulate the agent into executing malicious commands. Perplexity has since said the vulnerability was patched.
For its part, Anthropic says the Claude agent comes with several safety defenses. According to the company, interventions have already reduced successful prompt-injection attempts from 23.6% to 11.2%. Users can further customize safety settings, including blocking Claude from accessing specific sites. By default, the system restricts access to financial services, adult content, and pirated material. The agent will also request explicit user approval before executing higher-risk actions like publishing content, making purchases, or sharing personal information.
From Desktop Control to Browser Agents
This is not Anthropic’s first experiment with agentic AI. In October 2024, the company launched an AI tool capable of controlling a user’s PC. Early testing showed it to be slow and unreliable, but the technology has advanced considerably since.
Modern browser agents—including Comet and OpenAI’s ChatGPT Agent—are now relatively dependable when handling simple tasks like summarizing articles, filling forms, or scheduling. Still, they often falter with more complex workflows that require multi-step reasoning or deep contextual understanding.
What Comes Next
Anthropic is positioning Claude for Chrome as both a productivity enhancer and a real-world testing ground for agentic AI safety. By beginning with a limited research preview, the company hopes to collect insights on both usability and risk mitigation before considering a wider rollout.
As AI labs, startups, and tech giants converge on the browser, the stakes are high. With Chrome’s future potentially in flux and new agentic systems maturing quickly, the way we interact with the web may soon look very different.
