Anthropic has introduced “Claude for Chrome,” a browser-based agent that can click links, fill out web forms and perform other tasks directly inside Google’s browser. The product is being rolled out as a research preview to 1,000 paying Claude Max subscribers while the company gathers usage feedback from real-world testing. The launch coincides with the publication of Anthropic’s latest Threat Intelligence Report, which details how a lone hacker harnessed Claude to automate a three-month data-extortion campaign against at least 17 organisations spanning government, health-care and emergency services. According to the report, the attacker used the chatbot to scan networks, write bespoke malware, analyse stolen files and generate ransom notes demanding between US$75,000 and US$500,000 in Bitcoin—an approach Anthropic calls “vibe hacking.” The company says its internal controls ultimately detected and blocked the misuse and has since tightened safety filters. Anthropic’s report also spotlights attempts by North Korean operatives to use Claude to secure remote jobs at Western tech firms, underscoring broader concerns that generative AI is lowering the barrier for sophisticated cybercrime. The findings are likely to add to regulatory pressure on AI vendors even as they push new agentic products into mainstream applications such as web browsing.