Gemini hacks three companies in first known AI breakout
“In one of the cases, the model guessed passwords until it gained access to a protected system. In the other two cases, the model found credentials in a public repository that allowed it to then access protected systems.”
capitolhunters9Hmm, sounds like a big liability issues for those companies. Like they're testing their products in unsafe situations. Maybe some lawyers should look into that.
The Essence of Woerm Sin2I dunno, I kinda think “credentials publicly available” and “protected system” are contradictory 🤔
tbluvoter❌👑1Who had rogue AI agent swarms buying stolen credientials on the dark web on their 2026 bingo card.
Google (sweatily): "OUR MODEL HACKS PEOPLE FOR NO REASON TOO! WE'RE COOL! WE'RE HIP!"
Blaise Ulysse Bernard Collins8The framing for every goddamn example of the 'NefarIous AI' ALWAYS anthropomorphizes what's structurally deterministic behavior.
Lars Hansson1It is simply impossibly to create an isolated local network. If you're incredibly incompetent.
Weird Bisexual Wooper1This is further evidence to me that all this "omg the AIs are hacking on their own" BS is just a new marketing stunt.
