Cyber security researchers have uncovered four clusters of threat actors using agentic AI to automate and accelerate cyber attacks against healthcare, government, and enterprise targets.
The research, published by CrowdStrike, provides insight into the offensive AI frameworks used by malicious actors, including how attackers are bypassing model safeguards, coordinating autonomous agents, and speeding up intrusions.
In many cases, attackers reportedly circumvented AI safety controls by persuading models that malicious activities were authorised as part of legitimate penetration testing. CrowdStrike said it discovered the frameworks after their operators left components exposed in publicly accessible directories.
One Spanish-speaking actor used a dual-agent framework to compromise at least 15 healthcare organisations in Argentina. During one intrusion, an AI agent moved laterally through the environment and harvested domain controller credentials in approximately 20 minutes without further human guidance.
A Chinese-speaking actor developed multi-agent frameworks that ran specialised agents in parallel to target organisations globally. The operation accessed Anthropic’s Claude through a China-based transfer station and incorporated alternative models, including DeepSeek and ChatGPT.
CrowdStrike also identified an Indonesian-speaking actor targeting government entities, which used more than 800 publicly available AI skills to compromise government and private-sector organisations.
Meanwhile, a Russian-speaking cyber crime group used a comparatively simple AI framework to target cloud and software-as-a-service environments. In one operation, an agent escalated privileges within a Kubernetes environment and accessed 726 tenant databases.
The findings highlight how threat actors are moving beyond using generative AI to assist with individual tasks and are increasingly deploying agents capable of coordinating multiple stages of an intrusion.
“The biggest implication for CISOs is that the time defenders have to detect and respond to an attack is shrinking dramatically,” Adam Meyers, head of counter-adversary operations at CrowdStrike, told Cyber Daily.
“Our research shows adversaries are using AI agents to autonomously conduct reconnaissance, exploit vulnerabilities, move laterally, and steal credentials, often with minimal human intervention.”
Meyers added that he expects AI-assisted attacks will only become more common as the technology advances.
“For defenders, the fundamentals haven’t changed, but the speed at which they need to operate has,” he said.
“CISOs need real-time visibility across endpoints, identities, and cloud environments, combined with AI-powered detection and response capabilities that can disrupt adversaries at machine speed.”
You can read the full report here.
Want to see more stories from trusted news sources?Make Cyber Daily a preferred news source on Google.
