Google identified and stopped the first zero-day exploit it believes was developed entirely by AI. The exploit, a Python script that bypasses two-factor authentication on an open-source system administration tool, was discovered by Google’s Big Sleep AI agent before attackers could deploy it. Google worked with the vendor to patch the vulnerability and disrupted the planned mass exploitation event.
What Happened With the AI-Developed Zero-Day?
Google’s Threat Intelligence Group (GTIG) found evidence in the exploit’s Python code that an AI model was used to find and weaponize the vulnerability. Signs included “hallucinated” CVSS scores and “structured, textbook” formatting consistent with LLM training data. The exploit targeted a high-level semantic logic flaw where a developer had hardcoded a trust assumption into a 2FA system.
How Did Google Detect the Threat?
Google DeepMind and Project Zero’s Big Sleep agent — an AI system designed to search for unknown security vulnerabilities — found the flaw before the criminal group could exploit it. Google also introduced CodeMender, an AI agent that uses Gemini’s reasoning capabilities to automatically fix critical code vulnerabilities. The defensive AI found the flaw before the offensive AI could weaponize it in production.
What Does This Mean for the Future of Cybersecurity?
The incident marks a threshold: AI vulnerability discovery has crossed from research curiosity into operational reality. The GTIG report documents state-sponsored actors from China, North Korea, and Russia using AI for vulnerability research, autonomous malware using commercial AI APIs, and supply chain attacks targeting the AI software ecosystem.
Key Takeaways
- First confirmed AI-developed zero-day exploit detected and stopped
- Big Sleep AI agent found the vulnerability before criminals could exploit it
- Exploit bypassed 2FA on a popular open-source admin tool
- Python script showed telltale signs of AI generation
- Google disrupted the planned mass exploitation event
- GTIG report warns of industrial-scale AI-enabled hacking
Frequently Asked Questions
Was Gemini involved in creating the exploit? No. Google specifically states it does not believe Gemini was used. The AI model used by the attackers has not been identified.
Who was behind the attack? Google says “prominent cyber crime threat actors” with a strong record of high-profile incidents and mass exploitation were planning the attack.
How common are AI-developed exploits now? This is the first confirmed case Google has detected, but the company warns that criminal actors are increasingly using AI to find and exploit vulnerabilities at scale.