Google says AI ran a hack alone for 1st time
Google said hackers ran a cyberattack fully by AI for the first time, with no human at any step. Its Threat Intelligence Group shared the report on the 16th of November 2026.

Google spots a hack run by AIGoogle said hackers ran a cyberattack fully by AI for the first time, with no human at any step. Its Threat Intelligence Group shared the report on the 16th of November 2026.
Google tied the tool to Russian speakers, based on Russian notes in its code, CyberScoop reported. Google did not name a group or a country.
How the tool works
Google calls the system exploit_agent. It links several AI agents that compare popular software before and after a security fix.
When a fix comes out, the tool works out which flaw it closes. It then writes an attack and runs it on its own.
Who was hit
Data on the hackers' server pointed to French websites in media, entertainment and education, CyberScoop reported. Google also found scripts that guess passwords and plant hidden back doors.
Google also saw open attack tools called CAIRN and PentAGI used against a US aerospace group. Groups across Asia were also hit.
Below we break down the rise
Google counted 22 zero day flaws used in attacks in August alone. A zero day is a flaw the software maker did not know about.
That is twice the 11 per month seen earlier in 2026. It is almost three times the 2025 average of 8 per month, by our arithmetic.


What is confirmed and what is not
Google calls this the first fully automated attack it has seen. Outside experts have not yet checked that claim.
Key facts
Google shared the report on the 16th of November 2026.
It says an AI tool ran a full attack with no human.
The tool targeted French websites, Google found.
Google counted 22 zero day flaws used in August.
Source: Google Cloud Blog: GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold