🚨 Google found hackers' AI turning security fixes into attacks, with no human...

🚨 Google found hackers' AI turning security fixes into attacks, with no human involved When popular web software got a security patch, the AI compared the code before and after, worked out which hole had been fixed, wrote an exploit and ran it. Nobody stepped in at any stage. Google's threat intelligence team found the tool, which it calls exploit_agent, on an exposed server along with data stolen from victims. The targets looked like French media, entertainment and education sites. A note in its settings, written in Russian, told the operator to enter the AI keys "right here, not in the chat." It is the first time Google has seen attackers fully automate finding and exploiting flaws, its new report says. In another case, a suspected North Korean group broke into crypto firms and edited their Claude Code settings, so starting the app downloaded a backdoor. For now these tools are noisy and leave traces. Google expects them to get quieter in about six months. @aipost 🏴

