Google flags AI‑assisted zero‑day use

Cybersecurity people have been warning about this for a while. Not just AI writing phishing emails or cleaning up malware code, but AI actually helping find and weaponize new software flaws. Google now says it has seen that happen in the wild. The company’s Threat Intelligence Group said it disrupted a criminal campaign built around a zero-day 2FA bypass in a popular open-source web administration tool before the exploit could be used at scale. ### What’s the actual news? The new part is simple and pretty serious — Google says this is the first time it has identified a threat actor using a zero-day exploit that it believes was developed with AI. A zero-day matters because the vendor does not know the flaw exists yet, so there is no patch ready when attackers move. In this case, Google says the actor planned a mass-exploitation event, but the company’s proactive discovery and disclosure to the vendor may have stopped it. (blog.google) ### What was the exploit supposed to do? It targeted two-factor authentication. Google says the exploit was implemented in a Python script that let an attacker bypass 2FA on a popular open-source, web-based system administration tool. That is the kind of bug that turns a stolen or guessed password from “not enough” into “enough,” which is why this stands out. Google has not publicly named the tool, basically to avoid handing copycats a map. (blog.google) ### Why does Google think AI helped build it? Because the code looked weirdly like model output. Google pointed to “educational” docstrings, a clean textbook-style Python structure, detailed help menus, an ANSI color class, and even a hallucinated CVSS score embedded in the script. None of that proves a chatbot wrote every line, but taken together it is a strong forensic clue that an LLM helped generate or refine the exploit. That last part is an inference, but it is exactly the one Google is making. (cloud.google.com) ### Why is the 2FA angle such a big deal? Because 2FA is supposed to be the backstop. If an exploit can step around it, the attacker no longer has to intercept a code or trick a user in real time. The bug does the hard part for them. That makes a mass campaign much more scalable — one working bypass can be reused across lots of targets until the patch lands. (uk.pcmag.com) ### Is this just one weird case? Probably not. Google’s broader report says threat actors are already using AI for vulnerability research, malware development, defense evasion, synthetic media, and more autonomous operations. It also says actors linked to China and North Korea have shown interest in using AI for vulnerability discovery, while suspected Russia-linked activity has used AI-generated decoy logic and obfuscation in malware. (cloud.google.com) ### So are we in the “AI hackers everywhere” phase? Not quite — but the direction is clear. The scary part is not that AI magically invents unstoppable exploits. It is that it lowers the labor cost of the whole attack chain. Think less “robot super-hacker” and more “cheap junior staff for every bad actor” — code cleanup, bug hunting, exploit scaffolding, phishing text, infrastructure scripts, all faster than before. (blog.google) ### What should defenders take from this? Do not treat 2FA as a force field. Patch fast, reduce internet-facing admin surfaces, and assume attackers can now iterate on exploit code much faster than before. Google’s own M-Trends material already showed the handoff between initial access and follow-on operations collapsing to seconds in some investigations. AI plugs neatly into that acceleration. (blog.google) ### Bottom line? This is less about one unnamed tool than about a threshold getting crossed. Google says it caught the first AI-assisted zero-day exploit it has seen in the wild. If that call is right, the industry just moved from debating whether AI could help discover exploitable bugs to dealing with attackers who are already trying it. (blog.google) (security.googlecloudcommunity.com)