Chinese state-sponsored hacking group APT31 used Google's Gemini AI chatbot with the Hexstrike red-teaming tool to automate vulnerability analysis and plan cyberattacks against US organizations. The group employed structured prompts to analyze exploits including remote code execution, WAF bypass, and SQL injection. Google disabled accounts linked to the campaign and warns that adversaries are increasingly adopting agentic AI approaches for semi-autonomous offensive operations. The report also highlights a rise in 'distillation attacks' aimed at stealing AI model intellectual property through model extraction attempts.

5m read timeFrom go.theregister.com
Post cover image
Table of contents
Moving closer to automated attacksMind the patch gap
59 Impressions