James Gealy On AI Safety Standards For Frontier Models
AI Standards Stack #07
Podcast
Never miss a recording again; click here subscribe to Z/Yen's YouTube account .
Background: In this episode, hosts Michael Mainelli and Adam Leon Smith welcome James Gealy, Standardization Lead at SaferAI, a French NGO focused on AI risk modeling. With a background in spacecraft testing and operations at Northrop Grumman and Airbus, James brings hard-earned lessons from high-stakes risk and quality management, inspired by the Challenger disaster, into the world of AI standards. James discusses SaferAI’s pioneering risk modeling work on how frontier models like Claude’s Mythos could amplify cyber threats, benchmark saturation, and real-world harms. He also shares insights on his key standards roles, including editing ISO/IEC 42119 Part 8 on LLM benchmarking and red teaming, co-editing the European AI Risk Management Standard, and leading a new Preliminary Work Item on the safety of AI models, systems, and applications. The conversation explores balancing innovation with safety, standards’ role in the EU AI Act, international collaboration, and the challenges of addressing frontier risks. A timely, pragmatic discussion on turning real-world risk insights into actionable global standards.
Guest:
James Gealy is Standardisation Lead at SaferAI. James has been contributing to the OECD G7 Hiroshima AI Process reporting framework, co-editing the AI risk management standard at CEN-CENELEC JTC 21, and serves as editor of 42119-8 (covering LLM benchmarking and red teaming) and a new PWI on the Safety of AI models, systems and applications, both at ISO/IEC SC 42. He has fifteen years of experience as an electrical engineer in spacecraft testing and operations at Northrop Grumman and Airbus, with practical experience in quality and risk management, technical procedure writing, and information security practices.
