Google Says Russian-Speaking Hackers Ran an End-to-End AI Cyberattack

Summary
Google researchers say they identified an operation that automated vulnerability discovery and exploitation using AI agents. They attribute it to Russian-speaking hackers based on Russian-language code comments, but named no group or state.
Key points
- Google’s Threat Intelligence Group and Mandiant said researchers believe this was the first intrusion they observed running end to end without a human in the loop.
- The system, dubbed exploit_agent, used AI agents and language models to compare software before and after patches, infer vulnerabilities, draft exploits and run them automatically.
- Scripts also supported internal reconnaissance, password guessing and webshell installation; one script removed rival attackers’ webshells.
- Google said staging-server data pointed to targeting of French media, entertainment and education websites. The researchers named no group or state, citing Russian-language comments in code as attribution evidence.
- Google also reported CAIRN and PentAGI framework activity targeting an aerospace organization in the United States and organizations in Asia, and a China-nexus group developing an automated framework for likely espionage.
- After vendors patched two PHP vulnerabilities on Oct. 21, hundreds of purported exploits appeared within days, most reportedly nonfunctional. Google counted 22 zero-days exploited in August, compared with 11 per month earlier in 2026 and an average of eight in 2025.
Article Details
- Event Type
- Reported fully automated AI-driven cyberattack
- Impact
- Google said its researchers believe this was the first cyberattack run end-to-end by AI without a human in the loop. The reported system automated vulnerability discovery and exploit development and execution; investigators also found scripts for internal reconnaissance, password guessing, and installing webshells. Victim data on a staging server pointed to French websites in the media, entertainment, and education sectors. Google also reported separate activity involving offensive frameworks targeting a U.S. aerospace organization and organizations across Asia. No specific group or state was named.
MITRE ATT&CK
Products
Hugging Facemodel, and a summer incident in which an OpenAI agent hacked development platform Hugging Face, though itdid not appear aimed at specific targets.Mythosfindings arrive amid mounting concern over autonomous AI systems, including Anthropic’s April release of Mythos, a cybersecurity-focused model, and a summer incident in which an OpenAI agent hacked developmentPHPAfter vendors patched two PHP vulnerabilities on Oct.
Tools
CairnGoogle also said it observed open-source offensive frameworks, including CAIRN and PentAGI, in activity targeting an aerospace organization in the United States and organizations across Asia.exploit_agentThe centerpiece is a system the researchers dubbed “exploit_agent.” It chains together multiple AI agents with frontier and locally hosted language models to continuously compare code in widely used software before andPentAGIGoogle also said it observed open-source offensive frameworks, including CAIRN and PentAGI, in activity targeting an aerospace organization in the United States and organizations across Asia.
Countries
ChinaIt also cited a China-nexus group building an automated attack framework for likely espionage work.FranceVictim data on the staging server pointed to targeting of French websites in the media, entertainment and education sectors.United StatesGoogle also said it observed open-source offensive frameworks, including CAIRN and PentAGI, in activity targeting an aerospace organization in the United States and organizations across Asia.
Industries
AerospaceGoogle also said it observed open-source offensive frameworks, including CAIRN and PentAGI, in activity targeting an aerospace organization in the United States and organizations across Asia.EducationVictim data on the staging server pointed to targeting of French websites in the media, entertainment and education sectors.EntertainmentVictim data on the staging server pointed to targeting of French websites in the media, entertainment and education sectors.MediaVictim data on the staging server pointed to targeting of French websites in the media, entertainment and education sectors.
