Google Confirms Criminal Hackers Used AI to Discover Zero-Day Vulnerability
Published: 2026-05-12 Reading: 5 min Cybersecurity / AI Threats
On May 12, 2026, Google's threat intelligence division confirmed what cybersecurity experts have feared for years: criminal hackers have used artificial intelligence to discover and weaponize a previously unknown software vulnerability. This marks the first publicly confirmed case of AI-driven zero-day exploitation by a criminal group — a watershed moment for the security industry.
What Happened: The First AI-Powered Zero-Day
Google observed a group of "threat actors" planning a major operation that relied on a zero-day vulnerability they had discovered. The bug allowed them to bypass two-factor authentication on a widely used online system administration tool. Google declined to name the specific tool or the affected company.
A zero-day exploit targets a vulnerability that security engineers have had zero days to develop a fix for. What made this case extraordinary was the method of discovery: Google's analysis found evidence that the attackers leveraged an AI large language model — the same technology powering popular chatbots — to find the vulnerability.
"We have high confidence that the actor likely leveraged an A.I. model to support the discovery and weaponization of this vulnerability," Google's report stated.
John Hultquist, chief analyst at Google's threat intelligence arm, put it bluntly: "It's here. The era of AI-driven vulnerability and exploitation is already here."
How Google Responded
Google notified the affected company and law enforcement agencies and was able to disrupt the operation before it caused any damage. However, the company shared limited information about both the attackers and the target.
Key details Google did disclose:
- The vulnerability has been patched
- There is no evidence the attack was tied to an adversarial government
- Google did not reveal which specific AI model was used, only that it was most likely not Google's own Gemini or Anthropic's Claude Mythos
- The information was shared only with select companies and government agencies in the United States and Britain
Hultquist noted that groups tied to China and North Korea have been exploring similar AI-assisted techniques, suggesting this incident may be the tip of the iceberg.
The Mythos Factor: AI Models Built for Vulnerability Discovery
The timing of Google's announcement is significant. Just a month earlier, Anthropic unveiled Mythos, an AI model specifically designed for security vulnerability research. Mythos demonstrated remarkable capability in finding software flaws — so much so that Anthropic chose to share it only with a limited number of firms and government agencies in the US and UK rather than releasing it publicly.
The HN community raised important questions about the Mythos claims. Some users with access to the Cyber version noted that while Mythos is effective at cybersecurity work, it is only marginally better than its predecessor with the right jailbreaking. Others questioned whether Google's "high confidence" attribution was based on recovered LLM transcripts or simply inference from the vulnerability's characteristics.
Regardless of which specific model was used, the broader pattern is clear: AI models are becoming increasingly capable at code analysis and vulnerability discovery. Whether it's Mythos, GPT 5.5, Gemini, or open-source alternatives, the barrier to AI-assisted vulnerability hunting is dropping rapidly.
Why Criminals Benefit the Most
Hultquist highlighted a crucial asymmetry: compared with government spies who typically work slowly and quietly, criminal hackers have the most to gain from AI's speed advantage.
"There's a race between you and them to stop them before they can essentially get whatever data they need to extort you with, or launch ransomware," Hultquist said. "AI is going to be a huge advantage because they can move a lot faster."
This speed advantage manifests in several ways:
- Faster discovery: AI can analyze codebases at a scale and speed impossible for human researchers
- Rapid weaponization: Once a vulnerability is found, AI can help generate exploits faster than manual development
- Lower skill barrier: Attackers no longer need deep expertise to find complex bugs — the AI fills the gap
- Parallel exploitation: Multiple criminal groups using AI may independently discover the same vulnerability within hours of each other
The White House Rethinks AI Regulation
The incident landed amid a policy debate in Washington. After repealing former President Biden's AI guardrails, the Trump administration has been sending mixed signals about government oversight of AI.
Dean Ball, a senior fellow at the Foundation for American Innovation and former White House tech policy adviser, captured the tension: "I don't like regulation. I would prefer for things not to be regulated. But I think we need to in this case."
The challenge is finding a regulatory approach that addresses the cybersecurity risks of AI without stifling innovation. With models like Mythos demonstrating powerful vulnerability-finding capabilities, the question of who gets access to these tools — and under what conditions — becomes urgent.
What This Means for Cybersecurity in 2026
This incident signals several shifts in the cybersecurity landscape:
1. The Disclosure Timeline Has Collapsed
If AI can find zero-days in hours rather than months, the traditional 90-day disclosure window is dangerously long. Organizations need to assume that any vulnerability they're patching may already be known to attackers.
2. AI-Augmented Defense Is No Longer Optional
Defenders must use the same AI tools that attackers use. Automated code scanning, AI-powered patch triage, and real-time vulnerability monitoring are now essential, not nice-to-have.
3. Information Sharing Must Accelerate
Google's decision to share findings only with select US and UK entities highlights a tension: broad disclosure helps more defenders but also risks tipping off attackers. In an AI-accelerated threat landscape, faster and wider sharing may be necessary.
4. The Skills Gap Will Widen Before It Narrows
AI lowers the barrier for attackers but raises it for defenders who must understand and manage AI-driven threats. Security teams need to upskill rapidly or risk falling behind.
Key Takeaways
- Criminal hackers have crossed the AI threshold. This is no longer a theoretical risk — it's confirmed reality.
- Zero-day timelines are compressing dramatically. Plan emergency response assuming 48-hour independent rediscovery, not 90 days.
- AI vulnerability models like Mythos are game-changers. Their restricted distribution is a temporary measure, not a long-term solution.
- Defense must match offense. Organizations that don't adopt AI-powered security tools will be outpaced.
- Regulation is coming. The debate is no longer whether to regulate AI security tools, but how.
The first confirmed AI-driven zero-day attack is a warning shot. The technology that makes vulnerability discovery faster and cheaper for defenders does the same for attackers. The race has begun — and the era of AI-driven vulnerability exploitation is already here.