Anthropic has introduced Claude for Chrome, a research preview of its new browser-based AI agent powered by its Claude models. Initially, the rollout is limited to 1,000 subscribers on Anthropic’s Max plan, which ranges from $100 to $200 per month, with a waitlist now open for additional users. The tool allows users to interact with Claude in a sidecar window that retains browser context, while also granting permission for Claude to carry out tasks directly within the browser.
The launch signals Anthropic’s entry into what is fast becoming the next battleground for AI labs—AI-powered browsers. Perplexity recently released its own browser, Comet, equipped with a built-in AI agent, while reports suggest OpenAI is preparing to launch a similar product. Google, too, has rolled out Gemini integrations with Chrome in recent months. The competition is heating up further amid Google’s ongoing antitrust trial, where a judge may potentially order the tech giant to divest Chrome. This has already sparked interest from rivals—Perplexity even placed an unsolicited $34.5 billion offer for Chrome, and Sam Altman has hinted OpenAI might pursue an acquisition as well.
Alongside the innovation, Anthropic acknowledged the safety risks of granting AI agents browser access. The company highlighted indirect prompt-injection attacks as a concern, citing Brave’s recent discovery of vulnerabilities in Comet. In response, Anthropic has integrated several defenses, reporting that interventions reduced the success rate of such attacks from 23.6% to 11.2%.
The company has also built user safeguards, including the ability to restrict Claude’s access to specific websites and default blocks on categories such as financial services, adult sites, and pirated content. Additionally, Claude will always request explicit permission before performing high-risk actions, including publishing, purchasing, or sharing personal data.
This is not Anthropic’s first attempt at giving Claude agency over user environments. In October 2024, the company tested a PC-controlling AI agent, but early trials found it slow and unreliable. With rapid advancements since then, today’s agentic AI models—including Comet and ChatGPT Agent—are showing strong reliability for routine tasks, though they continue to face challenges with more complex workflows.
By launching Claude for Chrome as a research preview, Anthropic aims to identify and address emerging safety risks while refining its approach to agentic AI. As competition intensifies, the browser is quickly evolving into a critical interface for the future of human-AI collaboration.
