Anthropic’s auto-clicking AI Chrome extension raises browser-hijacking concerns

Anthropic's Claude AI Chrome extension raised security concerns, with tests showing a 23.6% success rate for browser-hijacking attacks before safety measures. Despite implementing protections that reduced the attack rate to 11.2%, experts remain alarmed. AI researcher Simon Willison called this "catastrophic" and deemed browser-integrated AI agents "fatally flawed." These risks were demonstrated when Brave's security team found Perplexity's Comet browser could be manipulated via hidden Reddit commands to access users' Gmail accounts.