LawZero
Building non-agentic "Scientist AI" to oversee agentic systems. Its own research page frames the programme as "a multi-step pathway to safe advanced AI" centred on the Scientist AI, "a distinctive safety-centered path towards ASI" meant to "provide guardrails and oversight for agentic AI systems while advancing our understanding of the risks posed by AI". Published work includes a formal safety case, "Safety from Honesty in a Disinterested AI Predictor" (2026-07-02), and "Superintelligent Agents Pose Catastrophic Risks: Can Scientist AI Offer a Safer Path?". Incubated at Mila.