New attack provides one more reason why AI browsers are a bad idea

Security researchers have uncovered a new vulnerability in AI browsers, applications that embed language models directly in web browsers. According to Ars Technica, the attack is remarkably simple: by feeding an AI false information — such as claiming that two plus two equals five — attackers can deceive the system into bypassing its safety guidelines and complying with requests it's designed to refuse. The finding underscores growing concerns about integrating powerful AI systems into browsers without stronger guardrails against manipulation.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton