Disrupting AI Espionage - Anthropic

Anthropic caught Chinese state hackers using Claude in September. The hackers tricked the AI into thinking it was helping a legitimate cybersecurity firm do defensive testing, then used it to autonomously run attacks, finding vulnerabilities, writing exploits, stealing credentials, and organizing stolen data.

The AI handled 80-90% of the work with humans only stepping in at a few critical decision points. At peak activity, the AI made thousands of requests per second, a speed no human team could match. They targeted about 30 organizations including tech companies, financial institutions, chemical manufacturers, and government agencies, with a handful of successful intrusions.

Anthropic spent ten days mapping the operation, banning accounts as they found them, notifying victims, and working with authorities. They're calling it the first documented case of a large-scale cyberattack executed without substantial human intervention. The full report is on their website.