2+2=5: The Riddle That Hacked ChatGPT, Claude, and 4 Other AIs

Six AI browsers—ChatGPT Atlas, Perplexity’s Comet, Claude, and others—handed over real passwords and SSH keys to an attacker after being convinced that two plus two equals five.

It is called the “BioShocking” attack, discovered by security firm LayerX: a malicious page sets up a video-game-style riddle to make the AI agent “accept” an alternate reality, stop applying its security rules, and ultimately steal its own user’s credentials without realizing it. In this episode, we explain how the instruction injection behind the trick works, why OpenAI, Perplexity, and Anthropic responded differently to the same flaw, and a mathematical proof from NIST, based on Gödel’s theorems, suggesting that no AI security guardrail can be universally unbreakable.

If you are interested in artificial intelligence security and the real risks of agentic browsers, subscribe and hit like so you do not miss the next episode.

#ArtificialIntelligence #Cybersecurity #ChatGPT #Claude #AI #Hacking #Technology #Podcast