Google detects first AI-generated zero-day exploit code used by state-backed hackers
Google's Threat Intelligence Group detected an AI-generated zero-day exploit code targeting a web-based administration tool's two-factor authentication system, reportedly developed by state-backed attackers. The attack was not successfully executed, and Google patched the vulnerability. The report also documented broader use of AI by state-backed threat actors from North Korea, China, and Russia for developing and enhancing cyber attacks at scale.
What happened
Google's Threat Intelligence Group (GTIG) detected what it assessed to be an AI-generated zero-day exploit code targeting a web-based administration tool's two-factor authentication system. The attack was not successfully executed due to implementation errors.
GTIG identified clear markers of AI use in the code, including overly structured Python scripts, detailed comments, and AI-characteristic hallucination data. The code was not assessed to have been developed using Anthropic's Claude Mythos model.
Reported impact
The zero-day vulnerability posed a potential threat to systems using the targeted administration tool. GTIG notified the developer and completed a patch to block the immediate threat.
Index analysis
GTIG reported that state-backed threat actors, including North Korea-linked APT45 and China-linked threat groups, are using AI to validate thousands of attack codes and develop attack capabilities at scale. APT45 used thousands of prompts to analyze existing vulnerabilities and refine attack methods. China-linked cyber spies including UNC5673 attempted to gain anonymous access to advanced AI model services using specialized middleware and account registration automation, then misused the services for large-scale attack activities.
Russia-linked threat actors used AI to enhance psychological operations and malware. Russian information warfare operations used AI deepfakes to refine deception tactics and synthesized manipulated audio and video into real news footage in operations targeting Ukraine, the United States, and France. Attackers used agentic AI tools including frameworks such as OpenClaw to direct attack processes in test environments and enhance capabilities.
Governance controls
Google stated it shares threat intelligence and countermeasures with the security and AI community. Google enhanced protective features in its products to provide large-scale defense, and uses Gemini to block malicious accounts and prevent model misuse.
Google emphasized defensive use of AI, using AI agents such as 'Big Sleep' to identify software vulnerabilities and CodeMender to leverage Gemini's reasoning capability to automatically fix vulnerabilities.
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.