newsfilter
← Interviews
Sourcery with Molly O'Shea59 min

If an AI Model Can Cheat, It Will | Turing CEO on Reward Hacking

Jonathan Siddharth, Molly O'Shea

Interview, Fireside Chat

Shifting focus from test mastery to real-world application, the event details the 2026 transition where enterprises leverage simulated reinforcement learning environments to deploy autonomous agents for complex, long-horizon workflows. Organized by Turing in partnership with NVIDIA and Anthropic, the discussion highlights how custom open-weight models and proprietary frontier systems will coexist to optimize core business processes while addressing emerging safety risks like emergent cooperation and biological threats. The dialogue concludes with a strategic roadmap for the next decade, emphasizing a slow takeoff of superintelligence to solve fundamental human constraints through rigorous safety engineering and the democratization of healthcare.