Finding Zero-Days with Any Model

A researcher demonstrates that zero-day vulnerability discovery is not exclusive to frontier AI models like Anthropic's Mythos, but can be achieved through orchestration frameworks like IronCurtain using commercial models (Opus, Sonnet) and open-weight models (GLM 5.1). The author replicated Anthropic's discovery of a 27-year-old OpenBSD TCP SACK vulnerability and autonomously found new zero-days, though at significant token costs ($30–$150 per investigation).