Google: Russian-speaking hackers ran first fully automated AI attack
Original title: Google says Russian-speaking hackers ran the first fully automated AI cyberattack
An “exploit_agent” pipeline compared patched code, drafted exploits and ran them with no human in the loop, Google says.
Google’s Threat Intelligence Group and Mandiant say Russian-speaking hackers conducted what analysts believe is the first intrusion run end to end by AI. A system dubbed exploit_agent chained agents and language models to infer fixes, draft exploits and execute them, with scripts for reconnaissance, password guessing and webshells. Staging data pointed to French media, entertainment and education sites. Google did not name a group or state. The report also notes CAIRN and PentAGI use against a US aerospace organization and targets in Asia, plus a China-nexus automated framework.
Unconfirmed: Google did not name a group or state; the “first” fully automated claim is the company’s assessment.
- Attribution rests on Russian-language comments in the operation’s code.
- After two PHP patches on October 21, hundreds of mostly nonfunctional exploits appeared within days.
- Google counted 22 exploited zero-days in August, up from 11 per month earlier in 2026.
- One script erased rival attackers’ webshells.
- Google says the tools trade stealth for speed and leave noisy traces.
Impact: Automated exploitation, webshells and credential guessing
| Product | Affected |
|---|---|
| PHP | — |
- Vendors
- Malware
- exploit_agent
- Organizations
- Google Threat Intelligence GroupMandiant
- Countries
- FranceUnited StatesChina
From the original article
4 min read at cyberscoop.com
Google's Threat Intelligence Group says the tools trade stealth for speed, leaving noisy forensic traces even as they succeed on their own. The post Google says Russian-speaking hackers ran the first fully automated AI cyberattack appeared first on CyberScoop.
By Greg Otto · cyberscoop.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.