Google's Threat Intelligence Group (GTIG) has reported finding an artificial intelligence framework that automatically studied security updates for popular web application software, worked out what flaw each update was fixing, wrote an exploit for it and then launched the attack, all without a human operator stepping in. The findings were reported by technology news site The Register on 16 November 2026.
According to the report, GTIG describes it as the first case it has observed of this kind of fully autonomous exploitation, and as a significant advance over the activity it saw in the third quarter.
How the tool worked
The framework, named exploit_agent, is described as a multi-agent system for vulnerability research and weaponisation, with Russian-language tooling. It can work with several frontier and locally run AI models, The Register reported.
Its method was to run continuous comparisons of web application code before and after security updates. When a software maker ships a fix, the difference between the old and new versions points to where the weakness was. The framework used that difference to identify potential vulnerabilities, develop exploits and execute them automatically.
What investigators found
The framework was discovered on an exposed staging server. The same server held data stolen from victim organisations and several scripts meant for use after a break-in: for scouting internal networks, guessing passwords by brute force, installing webshells and removing webshells planted by other attackers.
Google has linked the activity to a suspected Russian-speaking operator and has not named any group. Based on victim telemetry, the targets were French websites in the media and entertainment industries and in the education sector, the report said.
What is not known
The report does not name the web applications or the specific vulnerabilities involved, the AI models the framework relied on, how long the campaign ran or how many organisations were affected.
A wider pattern
The Register's report also lists other autonomous offensive frameworks that Google says are being used by criminals and espionage actors: ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI. PentAGI was used along with proxy services against the aerospace sector, it said.
"We have observed several threat actors separately experimenting with use of these frameworks and harnesses," the Google researchers wrote, as quoted by The Register.
The researchers pointed to the "low barrier to entry, ease of integrating new exploits, and massive scaling potential" of such tools, according to the same report.




