WORLDTECH NEWS Global technology intelligence.Contact
Cybersecurity SINGLE SOURCE

Google Says Hackers Are Handing Live Intrusions to Autonomous AI Agents

Urban commuters using turnstiles at a modern subway station, showcasing daily public transportation routine.
Illustrative photo.Photo by El gringo photo on PexelsGoogle logo shown for identification only; no affiliation with or endorsement of WORLDTECH is implied.

Google Threat Intelligence Group and Mandiant say criminal and state-linked actors have started turning live intrusions over to autonomous AI agents, a development the researchers call a turning point in adversarial use of AI. The account comes from the latest AI Threat Tracker update, as reported by The Cyber Express.

Attackers move beyond simple prompting

In its latest AI Threat Tracker update, the researchers note that actors in cybercrime, espionage and information operations are leaving basic prompting behind. They are moving toward autonomous multi-agent frameworks, automated exploitation pipelines and revenue schemes such as LLMJacking.

The group has observed several threat actors using open-source agentic pentesting frameworks during live intrusions, meaning against real targets rather than in test settings.

Named frameworks and the CAIRN example

Beyond ARTEX, which has drawn the most public attention, the researchers point to CAIRN, CyberStrikeAI and PentAGI, as well as experiments with HexStrike AI and custom tooling.

CAIRN operates through an OODA loop of observing, orienting, deciding and acting, in which agents coordinate via a shared blackboard. According to the findings, it was used against US aerospace targets.

Sources behind the findings

The report draws on Mandiant incident response work, the group's global tracking of threat actors and defenses running on Google's own platforms. Google is an Internet and software company based in Mountain View.

The account rests on one outlet's report of Google's update, so the scale of the activity and the number of affected organisations are not established by the available facts.

Open questions

  • How many organisations were targeted by the frameworks described in the update?
  • Which threat actors are behind the CAIRN campaign against US aerospace targets?

Key facts

  • Google Threat Intelligence Group and Mandiant describe adversarial use of AI as reaching a turning point.
  • Threat actors are shifting from simple prompting to autonomous multi-agent systems and automated exploitation.
  • The group has seen several threat actors use open-source agentic pentesting tools during live intrusions.
  • Named frameworks include ARTEX, CAIRN, CyberStrikeAI and PentAGI, plus experiments with HexStrike AI and custom tooling.
  • CAIRN runs an OODA loop in which agents share a blackboard and was used against US aerospace targets.

Sources & evidence