Hackers Used AI to Develop First Known Zero-Day 2FA Bypass

Source: The Hacker News — Ravie Lakshmanan, 2026-05-11

Summary

Google Threat Intelligence Group confirmed the first known in-the-wild use of AI-generated exploit code for zero-day vulnerability discovery and weaponization. An AI model produced a Python script exploiting a 2FA bypass in an unnamed open-source web-based system administration tool. This marks AI crossing from capability research into active offensive operations.

Vulnerability Details

  • Type: Two-factor authentication (2FA) bypass
  • Root cause: Semantic logic flaw from hard-coded trust assumption
  • Requirement: Valid credentials needed to exploit
  • AI fingerprints: Educational docstrings, hallucinated CVSS score, textbook Pythonic structure, ANSI color classes, detailed help menus — all characteristic of LLM training data

Google assessed with high confidence AI generated the exploit code.

Threat Actor Activity Documented

ActorAffiliationAI Usage
UNC2814China-nexusGemini for jailbreaking, embedded device research
APT45North KoreaThousands of repetitive CVE analysis prompts
APT27ChinaGemini for fleet management app development
Russia-nexusRussiaCANFAIL + LONGSTREAM malware via LLM-generated decoy code
TeamPCP/UNC6780UnknownTargeted AI development environments
UNC5673China-linkedCommercial tools + GitHub projects for LLM abuse
UNC6201China-affiliatedAutomated premium LLM account registration to bypass usage limits

PromptSpy Android Malware

  • Abuses Google Gemini to analyze screen content
  • Captures biometric data for authentication bypass
  • AppProtectionDetector module blocks uninstall
  • Dynamic C2 infrastructure updates for resilience

Shadow API Abuse (CISPA Research)

  • 17 shadow APIs provide unauthorized access to Claude and Gemini
  • Proxy services capture all prompts/responses for fine-tuning and knowledge distillation
  • Gemini-2.5-flash accuracy dropped from 83.82% (official API) to ~37% on medical benchmarks via shadow APIs

Key Quote

“AI is already accelerating vulnerability discovery, reducing the effort needed to identify, validate, and weaponize flaws… discovery, weaponization, and exploitation are faster.” — watchTowr Head of Threat Intelligence

Significance

First confirmed case of AI used offensively for zero-day discovery in the wild. Confirms threat intelligence warnings about AI compressing the exploit development lifecycle.