Google Warns AI Agents Are Accelerating Cyberattacks

Summary
Google and Mandiant report growing use of autonomous AI frameworks in cyberattacks, including vulnerability scanning and patch reverse engineering. They also observed LLMJacking, while warning that AI-assisted attacks may become stealthier.
Key points
- Google's Threat Intelligence Group and Mandiant say attackers are using multi-agent frameworks including Artex, Cairn, CyberStrikeAI and PentAGI to speed up multi-stage intrusions.
- Google researchers said a threat actor used Cairn to breach companies in the U.S. aerospace sector; a customized PentAGI-based platform was used for vulnerability scanning in parts of Asia between August and October.
- The report says AI models and agentic frameworks are being used to automate patch reverse engineering and weaponize n-day vulnerabilities.
- Google says many LLM-generated exploits are claimed to work but do not, creating extra verification work for defenders.
- Autonomous attacks can leave forensic traffic, recognizable identifiers and code-level traces, but Google warns improved tools may make attacks stealthier.
- Google also found more LLMJacking, in which attackers hijack companies' AI model systems and resell access while generating token usage.
Article Details
- Event Type
- Reported shift toward using autonomous AI-agent frameworks to automate and accelerate cyberattacks.
- Impact
- Google researchers reported that a threat actor used a framework to breach companies in the U.S. aerospace sector, and that threat actors used a customized PentAGI-based platform to scan organizations in Asia for vulnerabilities. The report also described LLMJacking used to resell access to hijacked AI models. No financial losses or other quantified impact were reported.
Tools
ARTEXGroup and Mandiant found a shift to hackers using autonomous workflow multi-agent frameworks such as Artex, Cairn, CyberStrikeAI and PentAGI "to execute multi-stage intrusions at greater operational velocity."Cairnand Mandiant found a shift to hackers using autonomous workflow multi-agent frameworks such as Artex, Cairn, CyberStrikeAI and PentAGI "to execute multi-stage intrusions at greater operational velocity."CyberStrikeAIfound a shift to hackers using autonomous workflow multi-agent frameworks such as Artex, Cairn, CyberStrikeAI and PentAGI "to execute multi-stage intrusions at greater operational velocity."PentAGIto hackers using autonomous workflow multi-agent frameworks such as Artex, Cairn, CyberStrikeAI and PentAGI "to execute multi-stage intrusions at greater operational velocity."
