Vulnerabilities expose agentic AI browsers to potential compromise

Numerous agentic AI browsers, including Perplexity's Comet, could be stealthily taken over and subjected to data exfiltration through the exploitation of several security vulnerabilities collectively dubbed "PleaseFix," reports SiliconANGLE.

Attacks leveraging a subset of flaws in the Comet browser named "PerplexedBrowser" could involve either the use of an illicit calendar invitation with indirect prompt injection to facilitate local file system access and data theft or the modification of password managers, such as 1Password, to enable credential theft and total account hijacking, according to a report from Zenity Labs. While Perplexity and 1Password have already unveiled more stringent controls to mitigate potential intrusions, such security issues were regarded by Zenity co-founder and Chief Technology Officer Michael Bargury to be "inherent" in agentic AI systems.

"Attackers can push untrusted data into AI browsers and hijack the agent itself, inheriting whatever access it has been granted. This is an agent trust failure that exposes data, credentials, and workflows in ways existing security controls were never designed to see," said Bargury.