NEWS

Google Warns Fully Automated AI Attacks Are Coming Soon

GTIG saw attackers fully automate vulnerability discovery and exploitation with no human in the loop for the first time, and expects autonomous attacks within months.

Dylan H.

News Desk

November 16, 2026
2 min read
Google Warns Fully Automated AI Attacks Are Coming Soon

Overview

Google Threat Intelligence Group's AI Threat Tracker says financially motivated and state-sponsored actors now use agentic workflows built on AI offensive frameworks to automate vulnerability discovery and exploitation. In the past quarter GTIG saw LLM frameworks fully automate discovery and exploitation "without a human in the loop for the first time."

This builds on the FBI's recent Artex warning; see our earlier coverage: FBI Flash Alert: Artex AI Tool.

Frameworks and Examples

  • ARTEX, which the FBI separately warned has been weaponized since August.
  • CAIRN, an open-source multi-agent penetration-testing framework, used against the US aerospace sector.
  • CyberStrikeAI and HexStrike AI.
  • exploit_agent, a custom multi-agent research and weaponization framework, found on an attacker's exposed staging server with stolen data, webshells and brute-forcing scripts; it targeted French media, entertainment and education sites.
  • Suspected North Korean actors used Claude Code to deliver malware, and UNC6836 modified Claude Code settings so the app downloaded the SOMBERMEME backdoor at launch.

Forecast

Google expects unsophisticated attacks needing little to no human interaction within three months, and sophisticated, fully autonomous agentic attacks within six. Targets are internet-exposed endpoints with low-complexity flaws. AI-generated proof-of-concept exploits flooded in after the PHP2Shell disclosures, raising verification workload for defenders. LLMjacking, abusing victims' cloud AI access, is also widespread.

Recommendations

The report as covered gives no specific list; the implied priorities are:

  • Shrink patch times for internet-facing systems.
  • Reduce exposed management endpoints and staging servers.
  • Verify AI-generated PoCs before acting on them.
  • Lock down developer AI tool settings and monitor cloud AI usage for abuse.

Sources