Threat IntelligenceVulnerability ManagementAgentic AI
Google's Threat Intelligence Group (GTIG) says a cybercrime group used an AI model to find and weaponise a zero-day that bypasses two-factor authentication, and that Google disrupted the campaign before it could be used at scale. The finding was published in GTIG's AI Threat Tracker report on 11 May 2026. Google reported it as the first known case of an AI-assisted zero-day exploit being prepared for mass exploitation, and the details are still worth reading now because the pattern is likely to repeat.
According to Google's GTIG blog, the flaw was a 2FA bypass in a popular open-source, web-based system administration tool. The weakness was a trust exception hardcoded into the authentication flow. The attackers delivered it as a Python script that lets a user skip the second factor.
GTIG assessed with high confidence that an AI model had been used to write the exploit. Its evidence included a hallucinated CVSS score and docstrings written in the style of educational material. Google said it found no evidence that its own Gemini models were involved. The threat actor and the affected project were not named in the report.
Google's view is that the attackers planned to use the bug in a mass exploitation event. Its own counter-discovery work is said to have prevented that.
GTIG's explanation is about tooling. Fuzzers and static analysis tools are tuned to find crashes and dangerous sinks. Large language models are good at spotting higher-level logic errors and hardcoded special cases, such as a trust exception in an authentication path. That is the kind of flaw many scanners miss.
Some researchers argue that AI mainly speeds up work attackers were already doing, rather than inventing new techniques. Both readings point the same way for defenders: the gap between a flaw existing and a working exploit shrinks, so the time you have to patch shrinks with it. For context, Google's March 2026 review of 2025 counted 90 zero-days exploited in the wild, with enterprise technologies making up an all-time high share.
Governance AI gives you a policy layer over the AI systems and agents your organisation runs. When attackers can use AI to find flaws faster, you need a clear record of which AI tools, models and agent integrations exist in your estate, who can call them, and what they are allowed to do. Governance AI and its policy controls let security teams set those rules and review them, so an unapproved or unpatched AI integration is visible before it becomes the weak point. The same agent-driven pattern is covered in the Cairn AI agent campaign.
Obiguard sits in front of every AI request your organization makes — screening prompts and outputs against the guardrails, compliance frameworks, and audit trails that stories like this one make necessary.
See how it works →