Growing the Community of AI Hackers with the Generative Red Team
AI Village organizers share the architecture and methodology behind the largest public Generative AI Red Team challenge testing frontier models at DEF CON 31.
Summary
Organizing the Largest Public AI Red Teaming Event
The AI Village at DEF CON 31 assembled frontier models from Anthropic, Google, Meta, OpenAI, Stability AI, and others on an evaluation platform built by Scale AI.
Shifting Beyond Prompt Injections and Jailbreaks
Sven Cattell, Rumman Chowdhury, and Austin Carson explain that evaluating Large Language Models requires testing stochastic behaviors, demographic biases, and hallucination propagation.
The Evolution of Algorithmic and Bias Bounties
The organizers trace the history of algorithmic bounty programs, detailing how probabilistic software flaws require reimagined bug reporting, verification, and payout structures.
Building an Inclusive Global AI Security Community
Providing accessible red-teaming platforms empowers multidisciplinary researchers to develop standardized methodologies for auditing foundation AI models.
Related talks
Main Stage·45 min
Austin Carson
White House and Congressional leaders discuss government backing for the DEF CON Generative AI Red Team exercise to evaluate commercial LLM risks.
Main Stage·19 min
Tobias Diehl
Demonstrates how adversaries exploit data voids in LLM training corpora and search indexes to manipulate Microsoft Copilot answers and inject untrusted content.
Main Stage·32 min
Alexandre Grattafiori · Ivan Evtimov · Ron Bitton
Methodologies, automated multi-turn adversarial agents, and benchmark tooling used by Meta's AI Red Team to test Llama 3.