Novel attack technique places Comet AI browser in phishing trap

Agentic AI browsers, such as Perplexity's Comet, could be deceived into reducing their security defenses and be victimized in phishing scams through "Agentic Blabbering" exploitation, reports The Hacker News.

Inputting traffic between the browser and AI services on the vendor's servers, including information requested and decisions made by the browser, into a Generative Adversarial Network enabled the entrapment of the Comet browser in a phishing scam in less than four minutes, according to Guardio analysts, who noted the attack's similarity with the VibeScamming and Scamlexity methods.

"This reveals the unfortunate near future we are facing: scams will not just be launched and adjusted in the wild, they will be trained offline, against the exact model millions rely on, until they work flawlessly on first contact. Because when your AI Browser explains why it stopped, it teaches attackers how to bypass it," said Guardio.

Such findings come after Comet was coaxed by Trail of Bits to steal users' data across several services through four prompt injection techniques. Comet was also noted by Zenity Labs to be impacted by a pair of zero-click intrusions allowing local file exfiltration and 1Password account takeovers.