An attacker used AI agents to steal thousands of credentials. The campaign took less than six hours to plan, build and execute.
Executive summary
Google’s September 8 threat report describes a human-directed attack using multiple AI agents. After compromising cloud infrastructure, the attacker built and ran a campaign that stole thousands of third-party credentials in under six hours.
What happened
Mandiant investigators found agents handling scanning and troubleshooting without manual intervention. The attacker supplied the objective and instructions.
This is a newly published account of activity from Q2 2026, not an attack that happened this week.
The 2050 museum label
The criminal enterprise finally solved its staffing problem.
Human-directed crime, not spontaneous AI rebellion. Google has not observed fully autonomous zero-day discovery and intrusion pipelines used by threat actors against real targets. Its report describes particular investigations, not how common this is across all cybercrime.
Original reporting, primary documents and public posts. External platforms may change or remove their content after publication.
September 8; Mandiant investigations, observed activity and limits
September 8 coverage of Google’s report
September 8; separates observed attacks from experiments
Dispatches from 2050
We’ll email you new exhibits, with the story, the sources and the museum’s commentary.