The Changing Physics of Cybersecurity and Drug Development with OpenAI President Greg Brockman

Summary · 2 min read

Greg Brockman is the president and a co-founder of OpenAI. He discusses the company’s work on frontier models, AI agents, cybersecurity and scientific discovery.

Brockman argues that AI can now produce new knowledge, citing nearly 400 important open math problems that OpenAI says its models have resolved, while noting that biology will depend on access to data. He sees drug discovery as the clearest scientific application, but says labs, simulations and regulatory processes remain bottlenecks. He describes proactive agents such as Dots and expects workplace and consumer assistants to converge, with interfaces adapting to users’ needs. On cybersecurity, he urges defenders to use frontier models to find and fix vulnerabilities, and argues that security spending may need to vastly outstrip software spending. He says OpenAI aims to keep humans in control and use AI to amplify human agency.

Minute by minute

1. 12:01

   Drug discovery is the clearest target. Brockman calls drug discovery the clearest scientific application of AI and argues that neural networks should be able to learn biology as they have learned language. He says AI could increase parallelism and shorten steps in the process. As one near-term possibility, he says a treatment for some conditions may already exist in the literature but be difficult to find.

2. 19:06

   Work and consumer agents will converge. Brockman expects work and consumer assistants to converge, but says their main difference will be cost: workplace users may justify more compute, while consumer products need to serve people cheaply or for free. He says OpenAI is trying to unify its consumer, coding and enterprise efforts. He also sees AI lowering barriers for people starting companies.

3. 21:49

   Interfaces will adapt to each task. Brockman says the best interface may be no fixed interface, with people interacting through speech or tools such as Slack and iMessage. He notes that workplace products still need admin controls and security features. He also describes AI-generated interfaces, such as a one-off app for booking a flight or assembling a bicycle.

4. 25:55

   AI creates a window for defenders. Brockman calls the current period a defenders’ window: frontier models can help defenders before comparable capabilities are broadly diffused. He says OpenAI’s Daybreak program gives defenders access to models and that the company committed $1 billion to help frontline defenders. Internally, OpenAI paused projects for about 25% of its production engineers while models probed infrastructure; teams fixed the issues found.

Published Oct 15, 2026.