HASHJACK ATTACK USES URL '#' TO CONTROL AI BROWSER BEHAVIOR

Summary from the TLDR InfoSec newsletter

HashJack is a novel indirect prompt injection technique that embeds malicious commands in URL fragments (after the # symbol) to manipulate AI browser assistants such as Google Gemini, Microsoft Copilot, and Perplexity Comet. The attack takes advantage of AI assistants' handling of full URLs, letting threat actors weaponize legitimate websites without compromising them and execute malicious instructions that bypass traditional security controls, enabling credential theft, data exfiltration, and more. Microsoft and Perplexity updated their products by October-November, but Google considered the Gemini vulnerability as expected behavior and chose not to fix it, leaving users vulnerable to this emerging AI security threat.