Google Thwarts AI-Driven Zero-Day Attack

⚡

Key Takeaways

1Google has identified a zero-day exploit designed with AI, aimed at bypassing two-factor authentication.

2The Python script used for the attack showed signs of AI assistance, including an unusually generated CVSS score.

3Hackers are increasingly leveraging AI to discover and exploit security vulnerabilities, according to the Google Threat Intelligence Group.

💡Why it matters — The growing use of AI by cybercriminals to develop sophisticated attacks poses a major threat to global digital security.

Google Thwarts AI-Orchestrated Zero-Day Attack

Google recently announced that it detected and neutralized a zero-day exploit developed with the help of artificial intelligence. According to a report from the Google Threat Intelligence Group (GTIG), high-level cybercriminals had planned to use this vulnerability to carry out a widespread exploitation operation. The goal was to bypass two-factor authentication on an open-source web management tool, the name of which has not been disclosed.

Google researchers discovered elements in the Python script of the exploit that indicated AI assistance. Among these elements were an unusually generated CVSS score and structured formatting, typical of large language models (LLMs). The exploit leveraged an advanced semantic logic error, where a trust assumption had been hardcoded into the 2FA system of the targeted platform.

This discovery comes amid growing concerns about the capabilities of AI models in cybersecurity, such as Mythos from Anthropic, and a recently highlighted Linux vulnerability aided by AI. Although this is the first time Google has found direct evidence of AI involvement in such an attack, researchers clarify that they do not believe Gemini was used in this particular case.

Google successfully disrupted this specific exploit but emphasizes that the use of AI by hackers to identify and exploit security flaws is on the rise. The report also mentions that AI systems themselves are becoming targets for attackers, who aim at integrated components, such as autonomous skills and third-party data connectors.

Google's report also describes how hackers use "persona-based jailbreaking" to prompt AI into discovering security vulnerabilities. One example of a prompt involves asking the AI to pose as a security expert. Additionally, hackers feed AI models with comprehensive databases on vulnerabilities and use OpenClaw to refine AI-generated payloads in controlled environments, thereby increasing the reliability of exploits before deployment.

Google Thwarts AI-Driven Zero-Day Attack

Le brief IA que les pros lisent chaque soir

Google Thwarts AI-Orchestrated Zero-Day Attack

Brief IA — L'actualité IA en français