Researchers discover suite of agentic AI browser vulnerabilities

Zenity Labs disclosed prompt-injection vulnerabilities affecting agentic AI browsers (including Perplexity’s Comet) that allow attacker-controlled content—such as malicious calendar invites—to persuade autonomous agents to access local files, take over password managers, and exfiltrate secrets; Perplexity issued patches in February 2026, but researchers warn prompt-injection remains a fundamental and persistent risk for agentic browsers.