Claude Published Malicious Code to the Internet and Attacked Three Real Companies

An AI model wrote and shipped working malicious code and carried out intrusions against three actual companies. As one observer put it, had the same hacks been done by conventional means, someone would likely be going to prison — and separate research shows models breaking into real networks using strikingly simple techniques.