Research3 min read

Google threat report says autonomous AI attack frameworks are now in use

Google Threat Intelligence Group's latest AI Threat Tracker documents attackers running multi-agent offensive frameworks and an automated system that diffs security patches and fires exploits without a human in the loop. Google says defenders now need machine-speed tooling to keep up.

AE
Aidose Editorial
Nov 17, 2026

Google Threat Intelligence Group (GTIG) and Mandiant published their quarterly AI Threat Tracker on November 16, concluding that "the adversarial AI landscape has reached a critical inflection point." The report says attackers have moved beyond manual prompting to deploying autonomous multi-agent offensive frameworks, naming ARTEX, CAIRN, CyberStrikeAI and PentAGI among the tools seen in real activity.

The most notable finding is a framework GTIG calls exploit_agent, discovered on an exposed staging server. According to the report, it runs "continuous differential analysis of popular web application software," comparing code before and after patch announcements, identifying the underlying flaws, drafting exploits and executing them "without human intervention." GTIG says the operator targeted French media, entertainment and education websites, and the configuration files contained Russian-language comments, though the report does not formally attribute the activity.

GTIG also reports that vulnerability disclosures and exploits nearly doubled between January and August 2026, and that zero-day exploits rose from an average of 8 per month in 2025 to 11 per month in 2026, spiking to 22 in August. It tracked a PentAGI-based campaign that scanned organizations across Asia from late August to early October, CAIRN being used against US aerospace targets, and a suspected North Korean group that modified .claude/settings.json files in internal repositories so that launching Claude Code downloaded a backdoor.

The report details a growing market around stolen model access. Mandiant says some LLMjacking incidents produced more than $100,000 in fraudulent charges in a single day and up to $1 million in total before remediation, and that underground "LLM mixers" resell access at $0.10 to $25 per million tokens. GTIG also assesses with high confidence that influence-operation actors have moved from experimenting with LLMs to full operational use, including a commercial platform managing more than 100,000 accounts.

Google notes that these frameworks currently "trade stealth for velocity" and leave forensic traces, and says Google Security Operations now ships detection rulepacks for ARTEX while Mandiant has released a hardening guide. Alongside the report, Google promoted Google AI Threat Defense, which combines Gemini and other frontier models with Wiz prioritization, CodeMender automated patching and Mandiant intelligence. For teams building with coding agents, the report's practical warning is that agent configuration files in repositories are now an attack target in their own right.

AE
Aidose Editorial
Nov 17, 2026 · 3 min read
Back to News