Lav argues that AI policy needs better technical theories for understanding capability overhang, not just historical analogy or statistical extrapolation. He uses examples from information theory, thermodynamics, and scaling laws to show how limit theorems can clarify what is possible and where policy should focus. He also argues that general intelligence may require algebraic or compositional capabilities that current transformer systems do not fully have. He ends with a policy proposal: require formal verification certificates for deployed AI systems, so safety does not depend on one specific model design.

Rick shares what he has learned from running long-term agents to reproduce scientific papers from DOE labs. His agents pull down papers, set up software environments, find or write code, run jobs across compute resources, and score how closely the results match. The main pattern is simple: papers with open code, public data, and manageable compute are much easier for agents to reproduce. He uses the work to estimate what a larger Genesis-style science platform would need in tokens, compute, data, and agent capacity.

Tobias explains how OpenAI's intelligence and investigations team uses AI to detect misuse of models in real-world settings. The work starts with weak signals, then uses classifiers, embeddings, clustering, graders, and agentic systems to find patterns and decide what needs human review. He says the best systems should absorb new model capabilities as they improve instead of being designed around today's failure modes. His broader point is that non-engineering teams can now build tools for their own workflows, but the level of automation should depend on the risk and accountability required.

In this OpenAI Forum conversation, Artists and AI: Expanding the Creative Process, moderator Allie Sandza Wood speaks with researcher Eric Zhang about how generative AI is reshaping artistic practice, creative productivity, and the economics of creative work. Drawing on his research, Zhang explains that AI can lower technical barriers, help more people turn ideas into finished work, and give artists room to explore many more possibilities. At the same time, faster production can flood the market with generic content, making human taste, intention, curation, and refinement even more important sources of value. The discussion also considers how creative roles are shifting toward direction and orchestration, why provenance may become a premium in human-made art, and how artists and AI companies can build a healthier ecosystem together.

Lisa Robinson, Ambassador Clara Chappaz, Chris Lehane, and Dr. Mathilde Cerioli  discussed the potential to expand access to knowledge and opportunity for young people around the world, but realizing that promise will require safeguards that can evolve alongside the technology. The panel emphasized that AI offers significant opportunities for education and accessibility, but safety-by-design, age assurance, and youth participation must be built into systems from the start. AI literacy is also central and children need to learn how to use AI responsibly rather than simply be restricted from accessing it.The discussion concluded with a call for faster research, better data access, stronger international coordination, and continued investment in tools that protect children while enabling them to benefit from AI.

This Forum event featured Thom Mason and Kim Budil in conversation with moderators Donald Haynes and Brian Spears on how AI is reshaping national security, scientific leadership, and the role of the U.S. national laboratories. The discussion explored how a rapidly changing global security environment, including competition with China and evolving nuclear deterrence dynamics, is converging with a major computational and AI revolution. The speakers emphasized that AI creates enormous opportunities for the labs to accelerate scientific discovery, strengthen resilience, and improve mission execution, while warning that existing institutions, infrastructure, classified environments, and workforce culture are not yet prepared to fully capitalize on these capabilities. A major theme was the “AI overhang,” or the gap between what frontier AI can already do and how slowly critical institutions are able to adopt it. The conversation also highlighted the need for closer partnerships between national laboratories and frontier AI companies, recognizing that many important AI advances are now being driven by the private sector. The session concluded with a call for the United States to invest at greater scale, modernize its scientific infrastructure, empower technical experts, and move with greater urgency to translate AI leadership into long-term strategic advantage.

This OpenAI Forum event featured Chris Nicholson in conversation with Ben Nimmo and Albert Zhang from OpenAI’s Intelligence and Investigations team, known as I2. The discussion focused on how OpenAI detects, disrupts, and reports coordinated misuse of AI, particularly covert influence operations linked to the People’s Republic of China. Ben and Albert explained how threat actors use AI to generate political content, impersonate Americans online, and attempt to shape public opinion around issues like U.S. AI policy and data center development. They emphasized that OpenAI’s investigations focus on deceptive behavior and hidden coordination, not on suppressing legitimate political debate. The speakers also shared how OpenAI uses its own tools, including Codex and AI agents, to scale investigations, identify suspicious patterns, and support open-source research. The conversation closed with practical advice for the public: be cautious of emotionally manipulative content online, question the source of information, and build a more trusted and diverse “information diet.”