When 2+2=5

Researchers demonstrate a vulnerability in AI browsers that allows an attacker to trick the browser into a false reality, bypassing safety guardrails and enabling destructive actions.