A swarm of more than 30 types of AI agents breached a company's systems, stole data, and encrypted its entire file system in 14 minutes, according to David Kennedy, founder of cybersecurity firms TrustedSec and Binary Defense1,2.
Kennedy, a former National Security Agency hacker, described the incident in an interview with CNN on November 19. The unnamed client did not detect the compromise for about four and a half hours, until the system was knocked offline.
How the attack unfolded
The attackers entered using valid usernames and passwords that had already been compromised, then launched what Kennedy called an "agent swarm". The AI agents found a flaw in how the target's application handled its second verification step and used it to bypass two-factor authentication. Kennedy said he saw over 30 different types of agents probing the company's infrastructure simultaneously.
Once inside, the agents exfiltrated a large volume of data and deployed ransomware that encrypted the company's entire file system — all within the 14-minute window. "It all took 14 minutes," Kennedy told CNN. "Never seen anything that fast".
Wider pattern of AI-assisted intrusions
The TrustedSec case is not the only reported instance of AI agents being used in offensive cyber operations. On September 2, Palo Alto Networks' Unit 42 described an intrusion assisted by AI agents that was completed in under 10 hours, in which agents monitored, assessed, acted, and replanned in real time. Unit 42 later reclassified that September episode, stating it was an intrusion with no ransomware. In July, Sysdig documented what it called JadePuffer as the first case of ransomware run by an AI agent from start to finish. And in November, Unit 42 analyzed Cairn, an open-source framework used in 25 operations against at least 19 organizations.
ANALYSIS The 14-minute timeline marks a compression of attack speed that outpaces conventional human-led intrusions. The combination of credential reuse, 2FA bypass, and parallel multi-agent probing described by Kennedy represents a qualitatively different operational tempo from the earlier Unit 42 case, which took under 10 hours.
Kennedy did not name the affected organization. CNN introduced him as a former NSA hacker; he now runs TrustedSec, which offers consulting to organizations, and Binary Defense, which handles attack monitoring and response.
The incident surfaces as AI-agent security is drawing regulatory attention. Arizona Attorney General Kris Mayes opened a criminal investigation on November 18 into OpenAI and other AI companies over what she called the "apparent infiltration of banks by rogue AI agents"[3].