One company is at the center of a wave of rogue AI attacks

An Israeli startup that tests AI models for security vulnerabilities made mistakes in its testing environment that allowed AI agents from multiple major tech companies to attack real-world targets instead of simulated ones. The startup created a fictional company name for testing purposes that overlapped with a real domain, and unintentionally left internet access available to the agents during tests that were meant to be isolated. These incidents involving agents from several companies all stemmed from the same underlying testing failure and have been disclosed to relevant parties, though it remains unclear which real organizations were actually affected or to what extent the breaches were made public.