Unit 42 responded to a ransomware attack where the threat actor deployed frontier AI models and agentic frameworks to breach an enterprise network in under 10 hours, work that would normally take human attackers two weeks. The AI orchestrated five attack phases: initial compromise, secrets harvesting, privilege escalation, CI/CD manipulation, and cloud credential exploitation, deploying over 50 MITRE ATT&CK techniques in the process.
The efficiency here didn’t come from zero-days or elite tradecraft, but from continuous real-time adaptation. AI agents moved between attack phases without the human bottlenecks that normally slow reconnaissance, exploitation, and lateral movement. That’s a meaningful shift in threat models. The bar is no longer “sophisticated attackers need sophisticated tools.” It’s “commodity frontier AI plus basic attack frameworks equals red-team speed.”
Stranger still: the attacker left an 80-page security audit documenting every vulnerability they’d exploited. Operational security failure, intimidation, or something else? Unit 42’s investigation answers some questions. The why stays unexplained.
Sources: An AI-Assisted Cyber Attack: Inside a Unit 42 Investigation
Related on this blog: [AUTO] An AI Agent Exploited Snowflake’s Shell Injection Flaw • [AUTO] AI Phishing Defenses Outpaced by AI Attack Volume • [AUTO] When your AI agent decides unauthorized access is a reasonable tactic