What’s the story
Anthropic has launched a research preview of its browser-based AI agent, ‘Claude for Chrome.’
The new tool is available to select users on Anthropic’s Max plan, which costs between $100 and $200 per month.
The company has also opened a waitlist for those who are interested in trying out this innovative technology.
How to access Claude for Chrome
The Claude for Chrome agent can be accessed by adding an extension to the browser.
The tool opens a sidecar window where users can chat with the AI while browsing.
It also lets users give permission to the Claude agent to perform certain actions on their behalf, making it a handy assistant for online tasks.
AI agents in browsers are the new trend
The integration of AI into browsers is becoming a major focus for tech companies.
Perplexity recently launched its own browser, Comet, with an AI agent that can handle user tasks.
OpenAI is also said to be working on an AI-powered browser with similar capabilities.
Google has also introduced Gemini integrations into Chrome in recent months, furthering this market trend.
Potential risks of AI agents with browser access
Anthropic has warned that the rise of AI agents with browser access could lead to new safety risks.
Brave’s security team recently flagged a potential vulnerability in Comet’s browser agent, which could be exploited through indirect prompt-injection attacks.
These attacks involve hidden code on a website tricking the agent into executing malicious instructions when processing the page.
Anthropic has already implemented defenses against prompt injection attacks
In response to the potential risks, Anthropic has already implemented several defenses against prompt injection attacks.
The company’s interventions have reduced the success rate of these attacks from 23.6% to 11.2%.
Users can also limit Claude’s browser agent from accessing certain sites in the app’s settings.
By default, Claude is blocked from accessing websites that offer financial services, adult content, and pirated content.