Thousands of hackers will gather at DEF CON to participate in the Generative Red Team Challenge, the largest red-teaming exercise for AI models. The challenge, announced by the Biden-Harris administration and supported by various organizations, aims to test the vulnerabilities and weaknesses of large language models provided by companies like Google and OpenAI. The models will be evaluated on a platform developed by Scale AI, and the challenge will also incorporate a capture-the-flag style point system. The event will provide valuable feedback to AI companies, particularly regarding embedded harms and emergent risks. Students from overlooked institutions and communities will also participate in the challenge.
https://venturebeat.com/ai/legions-of-defcon-hackers-will-attack-generative-ai-models/