OpenAI President Greg Brockman on GPT-5.5 “Spud,” AI Model Moats, and Cybersecurity Risks

Key points

• Greg Brockman confirms that GPT-5.5 is the model nicknamed Spud, and describes it as a new class of intelligence and a step toward a new way of getting work done with a computer.
• Brockman says the most remarkable thing about GPT-5.5 is not better coding, which everyone expected, but crossing the threshold of usefulness for general applications like slides, spreadsheets, and browser-based computer use.
• OpenAI shifted over the past 12 to 18 months from focusing on benchmarks to focusing on real-world applications across functions like finance, sales, and marketing.
• Brockman frames the competitive moat as the machine that makes the machine: a repeatable, end-to-end co-design across pre-training, mid-training, reinforcement learning, and data collection, not any single technique.
• On open-source distillation, Brockman argues distillation does not simply reproduce the same capability in a smaller, faster model, or OpenAI would do it itself.
• Brockman describes OpenAI's business simply: rent, build, or buy compute and resell it with a positive operating margin, and says demand for intelligence outstrips supply.
• On cybersecurity, Brockman says OpenAI has invested in cyber safeguards for years, has expanded its trusted access for cyber program, and believes in iterative deployment and democratic access rather than either extreme.
• Brockman says agents should be given ramping responsibility together with governance, safety, observability, and oversight, pointing to Workspace agents with IT visibility as an example.
• Brockman predicts a compute-powered economy where more compute poured into a problem solves it faster, and says we are headed toward compute scarcity despite massive infrastructure investments.

Notes