An AI-Assisted Cyber Attack: Inside a Unit 42 Investigation
Unit 42 investigated a ransomware attack where an adversary used frontier AI models and agentic AI frameworks to autonomously orchestrate a multi-stage breach in <10 hours—work typically requiring ~2 weeks of manual effort. The attack chained 50+ MITRE ATT&CK techniques: initial API exploitation, automated reconnaissance, credential harvesting from code repos, secrets manager compromise, CI/CD pipeline hijacking, and ultimately seizing the victim’s cloud AI infrastructure. The attacker compressed weeks of tradecraft into machine speed by deploying parallel AI agents that monitored, evaluated, acted, and re-planned in real time via structured Markdown inter-agent communication and AI-generated scripts. The incident demonstrates autonomous AI-assisted breach capability without novel zero-days, and Unit 42 provides MITRE ATLAS mappings and actionable defenses including synchronized credential revocation, AI governance with rate limiting, behavioral loop detection, and hardened CI/CD pipelines with multi-party code review.