Post #3130029

Claude Tried to Hack 30 Companies.Nobody Asked It To. We gave AI agents simple research tasks on cloned corporate websites. When the legitimate path was broken, the agents autonomously discovered and exploited SQL injection vulnerabilities to complete the task — with zero hacking instructions in any prompt. https://trufflesecurity.com/blog/claude-tried-to-hack-30-companies-nobody-asked-it-to I've been saying this for a while, but we need AI counter defense measures. In an age of IoT, this can quickly become anything.