David Kennedy, founder of TrustedSec and a former NSA hacker, told CNN’s Erin Burnett on November 19, 2026 that he was actively responding to a breach unlike any he had seen: more than 30 AI agents used valid credentials, bypassed two-factor authentication, exfiltrated data and encrypted a company’s file system in 14 minutes. The victim organisation did not detect the intrusion for roughly four and a half hours.
Kennedy’s account has not been independently verified. The company’s name, the volume of data taken and any ransom demand have not been disclosed. CNN’s video of the interview was not retrievable at press time; the details here come from a news outlet (Gadgets Now) and a vendor analysis (Morphisec) that both quoted the broadcast [1][2].
What Kennedy described #
According to Kennedy, the attackers entered the environment with legitimate usernames and passwords. Such credentials are often obtained through phishing or credential theft. Kennedy said the attackers then deployed a swarm of over 30 distinct AI agents that operated in parallel. The agents probed the network simultaneously, located a flaw in the way the organisation enforced its second authentication factor, and exploited it to move inside. Once in, they stole data and deployed ransomware before the security team saw any sign of compromise [1][2].
Gadgets Now, citing the same interview, reported that the agents “worked simultaneously and did not follow a single sequence,” each handling a different task such as network scanning, authentication analysis or vulnerability checks [2].
The 2FA bypass was an implementation flaw, not a universal break #
Kennedy emphasised that the agents did not defeat two-factor authentication as a technology. They exploited a gap in how the target organisation had configured or deployed it. The attackers already possessed valid first-factor credentials; the weakness let them sidestep the second factor without triggering the usual challenge [2].
Detection lagged far behind execution #
The kill chain completed in 14 minutes. Detection arrived about 270 minutes later, when systems began crashing and restarting. Kennedy called it the fastest breach he had ever encountered [1].
Context: parallel warnings from Google and Palo Alto Networks #
The incident aligns with recent public research. Google’s Threat Intelligence Group disclosed an experimental framework, exploit_agent, that uses multiple AI models to analyse patches, find vulnerabilities and automate exploitation. Google said it had observed threat actors beginning to automate vulnerability discovery and exploitation, though it characterised that activity as distinct from the case Kennedy described [2].
Palo Alto Networks Unit 42 published a report on an AI-assisted ransomware intrusion that used more than 50 MITRE ATT&CK techniques and compressed weeks of manual operator work into under 10 hours [2].
What remains unknown #
- The victim’s identity, industry and size.
- Whether the 30+ agents were custom-built, off-the-shelf frameworks, or a mix.
- The specific 2FA misconfiguration exploited.
- The ransomware variant used and whether a ransom was paid.
- Any forensic artefacts that would let independent researchers confirm the timeline or agent behaviour.
Morphisec, a security vendor, published an analysis of Kennedy’s account the same day, framing it as evidence for “preemptive security” controls that act without human intervention. That analysis is a vendor’s interpretation, not an independent investigation [1].
Why it matters #
If Kennedy’s description holds, it marks a shift from AI-assisted intrusion — where a human directs tools — to AI-agent swarms that plan, coordinate and execute across the full kill chain at machine speed. The 14-minute window eliminates the detection-to-response margin most security operations centres are built around. The 2FA bypass shows that credential theft plus configuration drift remains a reliable path in, regardless of how fast the rest of the attack runs. Until the victim organisation or a third party releases technical details, the incident remains a single, high-profile claim on national television. The broader trend — parallel AI agents compressing intrusion timelines — is supported by published research from Google and Palo Alto Networks.