Observed incident
Google observes PentAGI abuse beyond scanning into intrusions and persistence
Reported
Published here
What happened
GTIG reports most observed PentAGI activity remained reconnaissance or scanning, while operational artifacts indicated some operations reached active intrusion and persistence across multiple organizations.
Parties and reporting
AI tools and systems: PentAGI, Unspecified cloud and local LLMs
Outcome
Google published defensive guidance. One customized deployment likely scanned Asian organizations from late August through early October; another targeted aerospace. Victim totals, final damage and containment are not disclosed.
Operating lesson
Constrain valid credentials and administrative interfaces; correlate rapid permission probing and persistence changes with automated containment.
Evidence and limits
This row covers the report’s PentAGI observations only. ARTEX/UNC7228 material overlapping Storm-0940 and ConnectWise is not counted again. Scanning and targeting do not establish compromise; predictions of fully autonomous attacks are not observed outcomes.
Sources
- GTIG/Mandiant original investigation
Primary source ·