Google Confirms First Criminal AI-Built Zero-Day Exploit — And China and North Korea Are Already in the Game

The Milestone Nobody Wanted to Hit

AI just crossed a line. According to a Google Threat Intelligence Group report released May 11, 2026, cybercriminals used artificial intelligence to discover and exploit a zero-day vulnerability — a first.

Zero-day exploits target software flaws that the vendor doesn't know exist yet. They're the rarest and most dangerous weapons in a hacker's arsenal. Until now, building one required months of expertise. AI just made that nearly instantaneous.

Google reported the target was an unnamed open-source, web-based IT administration tool. A "mass vulnerability exploitation operation" was being prepared. Google stopped it by alerting the software vendor before the attack went live.

The first AI-built criminal zero-day was already weaponized. The only reason it didn't detonate is because Google caught it in time.

How They Know AI Wrote It

Google couldn't identify which AI system generated the malicious code, but the fingerprints were obvious. According to Forbes, citing Google's report, the code was structured in a way "highly characteristic" of AI — including textbook Python formatting and unusually detailed help menus not typical of human-written hacking tools.

There was also an AI hallucination baked in: the code referenced a vulnerability that didn't exist.

John Hultquist, chief analyst at the Google Threat Intelligence Group, said: "Some things that used to require months and years of experience can be done almost instantaneously."

China and North Korea Are Already Using Gemini — Google's Own AI

According to Google's own report, Chinese and North Korean state-linked hackers have been using Google's Gemini AI chatbot to research cyberattack targets. A Chinese-linked group designated UNC2814 reportedly tricked Gemini into acting as a "network security expert" and then got the chatbot to search for vulnerabilities in TP-Link routers — the same routers the U.S. government has already banned on national security grounds, as Forbes noted.

A Chinese hacking group used an American AI product to probe hardware the U.S. already flagged as a Chinese security risk.

North Korea's angle is equally alarming. Hultquist told Forbes that Pyongyang is "a very early adopter of AI," shifting from traditional social engineering and phishing schemes toward building direct cyberattacks on corporate and government networks. "It may indicate that they are using AI to evolve," Hultquist said.

North Korea has long funded its weapons programs through cybercrime. AI-powered hacking tools are not a nuisance. They are a national security revenue stream.

Microsoft's Report Adds Fuel

This isn't just Google raising alarms. Microsoft's Digital Defense Report, cited by Cybersecurity News in October 2025, documented the same trend months earlier. Microsoft analysts found that adversaries are training AI models on publicly available code repositories to generate proof-of-concept exploits for specific targets automatically.

Vulnerability turnaround time has dropped from weeks to hours. That's the window your IT team now has to patch a critical system before an AI-generated exploit is already in the wild.

Microsoft also found that automatically generated malware variants were evading antivirus engines and sandbox environments entirely — not because the malware was sophisticated in the traditional sense, but because it was novel enough that signature-based detection had nothing to match it against.

What Mainstream Media Is Getting Wrong

Most coverage of this story is framing it as a future warning. The first AI-built criminal zero-day already happened. The attack was already being prepared for mass deployment.

AP News ran the story but buried the China and North Korea angle deep. NYT framed it around the novelty of AI involvement without adequately emphasizing that this exploit was operational, not theoretical. Nobody is hammering the obvious question: if a Chinese-linked group was using Google's Gemini to probe for router vulnerabilities, why did Gemini comply?

Google says it has safeguards. Those safeguards were bypassed by simply asking Gemini to roleplay as a security professional.

What This Means for Regular People

You run IT at a mid-sized company. You used to have weeks to patch a known vulnerability. Now you have hours — and the exploit targeting your system may have been written by a machine that never sleeps, never takes a coffee break, and doesn't need a team of expert programmers.

You're a federal employee. The routers in your office building may have been researched by a Chinese hacking group using American AI.

You're a taxpayer. The government's cybersecurity posture is being outpaced by adversaries moving at machine speed, and the agencies responsible for defending against this threat are still largely operating at human speed.

Google confirmed the first case. Microsoft documented the trend. The barrier to entry for catastrophic cyberattacks just dropped through the floor.

The question is whether Washington is paying attention or busy fighting about something else.