Google's threat intelligence unit says attackers let AI agents run live intrusions

Threat actors have started handing whole stages of live intrusions to AI agents, and at least one operation no longer needs a person to turn a newly released patch into a working attack, according to Google Threat Intelligence Group (GTIG) and Mandiant.
What the researchers describe
Google Threat Intelligence Group (GTIG) and Mandiant say that attackers have begun passing entire phases of active intrusions to AI agents, meaning AI that works through multi-step tasks instead of answering a single question.
The same findings point to at least one operation that has reached the stage where no person has to step in to convert a freshly released patch into a functioning exploit.
Who is behind the findings
The findings come from GTIG and Mandiant. Google, an Internet and software company based in Mountain View and founded in 1998, is the parent of the threat intelligence group.
So far the account has been reported by a single outlet, and the available material contains no independent confirmation of the wider details of the operation.
Open questions
- Which threat actors run the operation that works without a human?
- How many intrusions have used AI agents in this way?