Google says an AI agent turned patches into exploits
A suspected Russian-speaking operator ran an AI agent that built and ran exploits with no human help, according to Google's threat intelligence team.

The facts
- According to The Register, Google Threat Intelligence Group (GTIG) says a suspected Russian-speaking operator used an AI agent to turn security patches into working exploits with no human in the loop, which it calls a first.
- Google found an exposed staging server hosting a multi-agent framework called exploit_agent. It compared code before and after security updates, found flaws, built exploits and ran them automatically.
- The server held data stolen from victim organisations and several post-compromise scripts. Victim telemetry suggests the operator targeted French media, entertainment and education websites.
- The report also names other AI offensive frameworks used by criminals and suspected espionage groups, including CAIRN, CyberStrikeAI, HexStrike AI and PentAGI. Google says it is highly likely more attackers will use them.
Case notes · Danny's take
Why it matters
A patch is meant to close a hole. Here, the patch became the map. An agent compared code before and after an update, found what changed and built an attack from it. The window between a fix being released and attackers using it keeps shrinking. If you run a business or a site, 'we'll update next month' is now a risk decision, not a schedule. This is no longer one skilled hacker typing prompts. It's tooling anyone can pick up.
For traders
Watch for confirmed breaches at companies running popular web software, and for how they disclose them. Treat early claims as unconfirmed until the firm or a regulator says so. If you hold tokens tied to exchanges or wallet services, check how fast those teams patch.
For builders
Patch speed is now part of your security. Automate updates for web software, and assume a released fix will be reverse-engineered quickly. Defenders can use the same kind of tooling to test their own code before attackers do.
Not financial advice. Do your own investigation.
This story was summarised with AI from The Register's reporting and checked against our editorial policy. Danny's take reflects his views. Disclaimer.
