
Google Warns of Increasingly Autonomous AI-Powered Cyberattacks
- Google's Nov. 16 threat report says attackers increasingly use autonomous AI tools to find vulnerabilities, write exploits and run attacks.
- Researchers found exploit_agent, a framework that discovered and exploited flaws with no human involvement in those stages.
- A suspected North Korean cluster targeted crypto firms in North America and Europe with AI-made fake recruiter profiles.
Google warned on Nov. 16 that cybercriminals are increasingly using autonomous AI tools to identify vulnerabilities, develop exploits and execute attacks with less human involvement.
Its latest threat intelligence report documents automated attacks against websites and new methods targeting cryptocurrency companies through compromised AI development tools.
Google Finds AI System Automating Vulnerability Exploits
In its Nov. 16 AI Threat Tracker, Google Threat Intelligence Group (GTIG) cleared that the attackers had moved beyond using chatbots for individual tasks toward systems that coordinate multiple stages of cyberattacks.
Researchers discovered an exposed server hosting a framework called exploit_agent. The system used multiple AI models to examine software updates, identify potentially exploitable vulnerabilities, generate attack code and execute it without human intervention.
Google cleared that the evidence from the server indicated an interest in French media, entertainment and education websites. Researchers also found tools associated with activities after a compromise, including internal reconnaissance and installing malicious webshells.
The discovery marked the first time GTIG had identified attackers using AI frameworks to fully automate vulnerability discovery and exploitation without human involvement in those stages.
AI Attack Frameworks Spread Across Multiple Industries
Google and Mandiant observed attackers using tools including ARTEX, CAIRN and PentAGI to automate parts of live intrusions. Some frameworks can scan targets, generate attack code and adjust their actions based on results.
GTIG observed CAIRN in activity targeting organizations including the US aerospace sector. It also tracked a customized PentAGI-based system conducting reconnaissance and vulnerability scans across organizations in Asia between late August and early October.
The report distinguished these observed activities from fully autonomous attacks across every stage of an intrusion. Google expects attackers to reduce human involvement further as the technology improves.
Researchers also noted a limitation: automated tools can generate recognizable network traffic and leave forensic evidence, making some operations easier to detect.
Cryptocurrency Firms Face AI-Assisted Recruitment Attacks
Google also documented activity attributed to a suspected North Korean threat cluster, MIDNIGHT NEPTUNE, targeting cryptocurrency organizations in North America and Europe.
The attackers created fraudulent recruiter profiles using AI-generated images and directed targets to fake meeting websites. Those sites prompted victims to execute scripts that could compromise their systems.
Following initial access, the attackers modified Claude Code configuration files inside internal repositories. The changes could trigger downloads of a backdoor named SOMBERMEME when developers launched the coding assistant.
The report also described attackers disguising GitHub access tokens with names resembling Microsoft Copilot. Google has responded to observed AI misuse by disabling associated accounts and projects, strengthening model safeguards and sharing threat intelligence with defenders.










