Google flags AI‑developed zero‑day

Cybersecurity people have worried for a while that AI would eventually help attackers find real software flaws faster than defenders could patch them. That worry just got a lot less theoretical. Google’s Threat Intelligence Group said on May 11 that it identified a threat actor using a zero-day exploit it believes was developed with AI, and that Google’s own counter-discovery may have stopped a planned mass exploitation campaign before it started. ### What is the actual news here? A zero-day is a software flaw nobody has publicly fixed yet, which makes it one of the most dangerous tools in hacking. Google is not saying “attackers used ChatGPT for emails” or “AI helped with phishing copy.” It is saying the attacker appears to have used AI in the much more serious part of the chain — finding and developing a working exploit for an unknown vulnerability. Google framed this as the first time its team has identified that kind of case. (cloud.google.com) ### Did Google say what got hit? Not really. Google said it reported the issue to the affected company, and that company patched the vulnerability, but the public write-up does not name the vendor or product in the summary that Google posted. What Google did say is that the actor planned to use the exploit in a wide-scale or mass exploitation event, which is the detail that makes this feel less like a lab curiosity and more like a near-miss. (cloud.google.com) ### Why is “AI-developed” such a big jump? Because the hard part in offensive security is not just writing malware. It is spotting a subtle weakness, proving it is exploitable, and turning that into something reliable enough to use in the wild. If AI can help with that pipeline, then the time between “bug exists” and “attack works” can shrink fast. Google’s own threat team has been warning that AI is moving from scattered experimentation into industrial-scale use inside attacker workflows. (blog.google) ### Is this just one criminal group? No — and that is the bigger pattern. In the same report, Google said China- and North Korea-linked actors have shown strong interest in using AI for vulnerability discovery. It also said suspected Russia-nexus operators used AI-driven coding to speed up infrastructure suites and polymorphic malware, including obfuscation networks and decoy logic tied to campaigns against Ukrainian targets. (cloud.google.com) ### Where does APT45 fit in? APT45 is a North Korean group Google and Mandiant have tracked for years. Google’s new report says North Korean actors, including APT45 in the broader trendline cited by outside coverage, are using AI to refine and scale cyber operations. That fits the group’s history — espionage, financially motivated operations, and attacks on critical infrastructure — but AI gives that playbook more speed and reach. (cloud.google.com) ### Why now? Because zero-days were already not rare. Google tracked 90 in-the-wild zero-day vulnerabilities in 2025, up from 78 in 2024, and nearly half targeted enterprise technologies. That means defenders were already dealing with a high-volume, high-value exploit market before AI started showing up in the discovery and weaponization layer. ### Does Google have a defense story too? (politico.com) Yes — and it is basically the mirror image of the threat. Google says it is using AI agents like Big Sleep to detect vulnerabilities and CodeMender-style systems to help fix them automatically. The company is making the obvious point: if attackers are using models to compress offensive work, defenders have to automate faster too. (cloud.google.com) ### So what changes now? The main change is psychological, but that matters. Security teams can no longer treat AI-enabled exploit development as a future scenario. Google is saying it has now seen one in the wild, or close enough to the wild that it moved to stop it. The bottom line is simple — the race is no longer about whether AI will matter in zero-days. It is about which side operationalizes it faster. (cloud.google.com) (blog.google)