Google disrupts AI-generated zero-day exploit attempt
Google's Threat Intelligence Group reportedly discovered and thwarted the first known attempt by cybercriminals to use AI to build a working zero-day exploit targeting a two-factor authentication system. The exploit showed signs of AI generation, including educational docstrings and a hallucinated CVSS score, and was intended for a "mass exploitation event" before being disrupted.
What happened
Cybercriminals used AI to build a working zero-day exploit targeting a two-factor authentication system, according to research from Google's Threat Intelligence Group (GTIG).
The exploit exhibited signs of AI generation, including abundant educational docstrings, a hallucinated CVSS score, and textbook Python formatting characteristic of LLM training data.
A threat actor group planned to deploy the zero-day in a 'mass exploitation event' but was thwarted by Google's 'proactive counter discovery.'
Reported impact
The vulnerability allowed bypassing two-factor authentication to access a popular online system administration tool, which could have enabled widespread unauthorized access.
Google stated the discovery represents a significant moment in the use of AI for malicious purposes and marks the beginning of an era where AI-driven vulnerability discovery has become operational.
Index analysis
The zero-day was classified as a two-factor authentication bypass stemming from a high-level semantic logic flaw where developers hardcoded a trust assumption, the type easily identified by large language models.
Researchers deduced that the AI model's reasoning capabilities allowed it to identify developer intent and semantic flaws that traditional vulnerability detection tools like fuzzers and static analysis typically miss.
Governance controls
Google's proactive threat intelligence monitoring enabled detection and disruption of the exploit before it could be deployed at scale.
Google noted it does not believe its Gemini model was used, but said it had 'high confidence' another AI model was responsible for generating the exploit.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.