Open-closed model gap, evals in the biorisk hierarchy, DARPA AI Forge, refusal benchmarks, OpenAI’s Rosalind Biodefense
Five Things (June 5, 2026): AIxBio, and a Red…
Open-closed model gap, evals in the biorisk hierarchy, DARPA AI Forge, refusal benchmarks, OpenAI’s Rosalind Biodefense