Red-Teaming after Mythos — Zico Kolter & Matt Fredrikson, Gray Swan
Gray Swan's Kolter and Fredrikson discuss AI red-teaming and indirect prompt injection after US export controls on Mythos and Fable.
“the risks of jailbreaks and (industry term) indirect prompt injection are suddenly the talk of the town”
Latent Space interviews Zico Kolter (OpenAI board, Safety & Security Committee) and Matt Fredrikson (Gray Swan CEO) on AI red-teaming and indirect prompt injection, spotlighted by US export controls on the Mythos and Fable models. Their work was cited in Anthropic's Mythos model card and Gray Swan's Shade tool was used to evaluate model robustness against prompt injection in coding environments. It matters because adversarial security has moved from niche research to a central regulatory and industry concern.