Google says hackers ran first fully automated AI cyberattack
Google's threat researchers say Russian-speaking hackers used an AI system that studied newly patched web software, wrote exploits and launched them with no human involved. Google calls it the first fully automated attack pipeline it has seen in the wild.
AI-assisted · US Brief desk · Sources listed below
What happened
In a quarterly report released Monday, Google Threat Intelligence Group and its Mandiant unit said they found an exposed staging server running a multi-agent hacking system they named "exploit_agent." It links several leading and locally run AI models. It compares popular web software before and after a patch, works out the flaw, drafts an exploit and runs it "without human intervention," the report says.
Why it matters
Tools like this start working the moment a fix is published, so the safe window for installing updates is shrinking, Google said. For anyone running a website or internet-facing server, that makes fast patching the main defense. Google said it is "highly likely" attackers will keep using such frameworks against organizations worldwide.
More context
From the report: "In the past quarter, GTIG identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time."
Who was behind it: Google did not name a group or a country. It tied the operation to Russian speakers through Russian-language comments in the code, CyberScoop reported, including a note telling the user to put AI keys for "auto-exploitation" in the file, "not in the chat." Victim data on the server pointed to French media, entertainment and education websites. The server also held stolen data and scripts for scouting networks, guessing passwords and planting webshells, including one that removed rival hackers' webshells.
The bigger trend: Google said attackers are also using open-source AI hacking frameworks such as ARTEX, CAIRN, CyberStrikeAI and PentAGI. Exploited zero-day flaws averaged 8 a month in 2025 and 11 a month in 2026, then jumped to 22 in August, the report said. After two PHP flaws were patched on Oct. 21, hundreds of claimed exploits appeared within days, most of them broken and likely AI-written, CyberScoop said.
How outlets framed it: CyberScoop led with "the first fully automated AI cyberattack." BleepingComputer's headline was more cautious, saying Google "warns that fully-automated AI attacks are coming soon." TechTarget focused on AI agents turning patches into exploits.
What's still unknown: Google did not say how many sites were breached, which software was targeted beyond the examples in its report, or whether the operator works for a government.
4 listed sources
- Google Threat Intelligence Group, GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, Nov. 16, 2026:
- CyberScoop (Greg Otto), Google says Russian-speaking hackers ran the first fully automated AI cyberattack, Nov. 16, 2026:
- BleepingComputer, Google warns that fully-automated AI attacks are coming soon, Nov. 2026:
- TechTarget, Google finds AI agents autonomously turning patches into exploits, Nov. 2026:
References listed by US Brief; a source count is not a verification score.
Editorial sourcing notes
Findings come from Google's own report; Google did not name the group, a country or the number of victims. The photo is a 2014 file image of a sign on Google's Mountain View, California, campus.
Like US Brief? You can support it with a tip.