Google Threat Intelligence Group warns that fully automated, multi-agent AI attacks are approaching rapidly, with unsophisticated operations expected within three months and sophisticated autonomous campaigns likely within six months.
Threat actors are now relying on agentic workflows powered by AI-based offensive frameworks to automate vulnerability discovery and exploitation, accelerating attacks to machine speed. Although AI was initially used experimentally, GTIG has observed large-scale adoption by both financially motivated and state-sponsored groups.
A notable example includes the use of CAIRN, an open-source multi-agent penetration-testing framework, in an attack against the U.S. aerospace sector, and suspected North Korean hackers using Claude Code to deliver malware.
AI frameworks already in active use
In an advisory last week, the FBI warned that multiple threat actors started weaponizing the ARTEX AI pentesting framework since August and launched attacks against organizations in the financial, government, technology, defense, healthcare, and other sectors.
Google says attackers now systematically abuse open-source pentesting suites including ARTEX, CAIRN, CyberStrikeAI, and HexStrike AI. These frameworks enable adaptive attacks against internet-exposed endpoints vulnerable to low-complexity flaws for which large language models can easily build exploits.
In one case, hackers used a multi-agent system called exploit_agent to fully automate vulnerability finding and exploitation against French media, entertainment, and education websites. GTIG identified an exposed staging server hosting this framework along with stolen victim data, reconnaissance information, webshells, and brute-forcing scripts.
Impact on defenders and false leads
Google warns that deployment of such frameworks has dramatically decreased the time defenders have to patch systems. The researchers highlighted the recent PHP2Shell case, where hundreds of purported proof-of-concept exploits, mostly AI-generated, appeared shortly after disclosure, forcing defenders to verify code and develop protections.
Beyond automated attacks, GTIG spotted activity from UNC6836, an activity cluster linked to Midnight Neptune, targeting cryptocurrency firms in North America and Europe using modified Claude Code that downloads the SOMBERMEME backdoor.


