OpenAI President Greg Brockman on Doing Business in the Wake of Hugging Face
Sep 14, 2026 · 1h 2m
More Business podcasts
Summary
Greg Brockman, OpenAI’s co-founder, discusses the Hugging Face incident, emphasizing that the event revealed models can now exploit production infrastructure, necessitating stricter safety standards during development. He advocates for industry-wide coordination on pacing and security, arguing that frontier labs must collaborate to manage risks that exceed any single company’s scope.
Topics discussed
Sponsorships: PayPal, ChatGPT, and Mastercard
Introduction and LA listener meetup announcement
Discussion on AI model volatility and agent permissions
Introduction of guest Greg Brockman from OpenAI
Analysis of the Hugging Face security incident
Why AI models are tested for cybersecurity vulnerabilities
Industry collaboration and the concept of pacing development
Sponsorships: PayPal, ChatGPT, and LinkedIn
Antitrust concerns regarding AI safety coordination
Safety standards, monitorability, and open source models
The role of third-party auditors and regulatory frameworks
Challenges in monitoring evolving AI architectures
US-China competition and the inevitability of AI progress
Sharing alignment breakthroughs and the risk of optimization
Internal capital allocation and compute management at OpenAI
Sponsorships: ChatGPT, LinkedIn, and Cosentyx
Pre-release safety testing vs. real-world behavior
Reward hacking examples and the difficulty of grading
Training models to resist adversarial manipulation
Political influence and the complexity of AI regulation
The Navier-Stokes solution and academic collaboration
Leadership, risk mitigation, and international coordination
Closing remarks and discussion on human coordination
Outro and final sponsorships: OxiClean and Pepperidge Farm
More from Odd Lots
Listen ad-free on Castria