I was honored to speak at IASEAI 2026 in Paris, including the opening panel hosted by Adam Gleave and a fireside chat with Brian Tse at the Open-Weight AI Risk Management workshop. I argued that openness does not automatically increase misuse risk, and that open-model safety needs platform- and ecosystem-level security features, not just model-level safeguards. It was also exciting to see strong interest in our Evaluating Evaluations work.