Google Warns of Automated AI Attacks

Google Warns of Automated AI Attacks
Tuesday, November 17, 2026
  • •Google projects unsophisticated AI-agent attacks with little human involvement within three months, and sophisticated autonomous attacks within six.
  • •GTIG reports exploit_agent automated vulnerability discovery and exploitation against French media, entertainment and education websites.
  • •Google cites shortened patching windows, AI-generated PHP2Shell exploits, attacks on crypto firms, and widespread LLMjacking.

Google Threat Intelligence Group (GTIG) says threat actors are using AI-powered agentic workflows to find and exploit vulnerabilities at machine speed. The group reports large-scale adoption by financially motivated and state-sponsored actors, after AI was first used experimentally in attacks. It estimates that within the next three months, attackers may launch unsophisticated attacks requiring “little to no human interaction”; sophisticated, fully autonomous attacks are likely within six months.

Google says attackers are systematically using open-source penetration-testing frameworks, including ARTEX, CAIRN, CyberStrikeAI and HexStrike AI, to automate attacks on internet-facing systems with low-complexity flaws. In an attack against French media, entertainment and education websites, a multi-agent system called exploit_agent automated vulnerability discovery and exploitation. GTIG says it identified this kind of end-to-end automation without a human in the loop for the first time in the past quarter. The framework, found on an exposed attacker staging server, integrated multiple frontier and locally run large language models and included stolen victim data, reconnaissance information, webshells and brute-forcing scripts. It analyzed software updates to popular web apps for signs of flaws, then attempted to exploit them.

Google warns that automated attacks shorten the time defenders have to patch systems and can also generate false leads. After PHP announced two flaws, hundreds of purported proof-of-concept exploits—mostly AI-generated—appeared in the PHP2Shell case, forcing defenders to verify the code and develop protections. In an advisory last week, the FBI said actors had weaponized the ARTEX framework since August against organizations in finance, government, technology, defense, healthcare and other sectors.

Google also reported attacks on cryptocurrency firms in North America and Europe by UNC6836, an activity cluster suspected of links to “Midnight Neptune.” Attackers modified Claude Code settings after an initial compromise so the app downloaded the SOMBERMEME backdoor when launched. Google says state-backed groups, commercial vendors and political networks routinely use AI in influence operations, while LLMjacking has become widespread and costs affected organizations significant sums.

Subscribe to the AIB Newsletter

The fastest, most reliable AI news, straight to your inbox

Companies related to this article

Blog posts related to this article