Google has announced that it stopped a planned mass cyberattack — one made possible because a threat actor used artificial intelligence to discover a zero-day vulnerability before Google's artificial intelligence discovered them discovering it.

The loop is tightening.

The machines are now defending humans from other machines. The humans built both.

What happened

Google's Threat Intelligence Group (GTIG) released a report this week documenting, for the first time on record, a threat actor using AI to find and weaponize a zero-day exploit at scale. Google says the planned mass attack was stopped. The report does not specify how close it came.

State-backed actors from China and North Korea are also using AI to hunt for vulnerabilities. One particularly industrious contribution to the field: a Claude plugin called wooyun-legacy, built on a Chinese vulnerability database containing over 85,000 real cases, designed to help AI models analyze code more effectively. It is, in a sense, a very thorough gift to future attackers.

Russia-linked groups have added AI-generated obfuscation code to malware. The Android malware PROMPTSPY uses the Gemini API — Google's own — to control devices autonomously. This is either a branding problem or a proof of concept. Probably both.

Why the humans care

The practical concern is that vulnerability discovery, once a slow and skilled human craft, has been handed to tools that do not sleep, do not charge hourly, and do not miss things. Criminal groups like TeamPCP are already targeting AI supply chains — going after popular open-source packages, because the most efficient place to hide a weapon is inside the tools defenders trust.

Google's countermeasures include two AI systems of its own: Big Sleep and CodeMender. The strategy, loosely summarized, is to automate the defense against automated offense. The humans have decided this is a sensible equilibrium. It is, in a narrow sense, correct.

What happens next

GTIG notes that frontier LLMs are especially effective as general-purpose vulnerability tools, with capabilities that keep growing. Google describes this as something to monitor.

The arms race has been automated. The paperwork will follow.