Revolutionizing Cybersecurity and AI-Driven Science with…

PodcastNo PriorsGreg BrockmanOct 15

Greg Brockman argues that AI could accelerate scientific and drug discovery, although data access and other bottlenecks will shape how quickly biology advances. He also describes a near-term opportunity for defenders to use AI to find and fix security weaknesses before more capable cyber tools become widely available. The discussion links these changes to the economics of AI compute, software security, and the possibility that clinical-trial expectations may evolve if AI proves better at predicting safety or efficacy.

Artificial Intelligence Cybersecurity AI Infrastructure

Preview · 15 more points in the app

Read the Full Summary

Key Takeaways

Brockman argues that as software becomes cheaper to produce, cybersecurity spending should rise relative to software spending because buyers still need reliable, secure systems. He also says the internet’s security debt and poorly audited software supply chains leave vulnerabilities that can affect multiple companies.

Brockman says defenders have a limited window to use frontier AI to secure systems before more capable cyber tools become broadly available, citing bank hacks in South Korea as an early warning. At OpenAI, about 25 production engineers used a leading model to probe infrastructure, and he says security testing will need to be repeated as models advance.

Brockman says AI could make formal verification more practical, despite its historical cost and complexity, and says OpenAI is training models to write secure code from the outset. Those approaches complement his account of using a leading model to identify and fix security issues in company infrastructure.

TECHNOLOGY

Cybersecurity Spending — Greg Brockman argues that cybersecurity spending should vastly outstrip software spending as software gets cheaper to produce. Buyers need reliable, secure, trustworthy software, he says, making resilience more valuable.

“I have a feeling that cyber spend, security spend should actually vastly outstrip software spend. Like, I really think that... Well, that's statement…”—

Mathematical Discovery — Greg Brockman says AI can accelerate scientific discovery, citing almost 400 important open mathematical problems that have been resolved. Mathematicians are reviewing the work for insights, though AI is not yet producing entirely new theories.

“Now there is still more to go…”—

Human Capabilities — Brockman argues that AI tools can expand what people are able to accomplish. He points to mathematics and asks how the same capabilities might apply to biology.

“It's really what can a human equipped with these tools achieve that was unthinkable before and I think it's almost like that the leverage of humans to do whatever you…”—

ECONOMY

Clinical-Trial Regulation — Brockman says drug development can take about 15 years, with successive trials growing more expensive and addressing different safety considerations before large-scale human trials. He argues regulatory expectations may need to change if evidence shows AI can predict safety or efficacy better, while noting that clinical-trial rules differ between Australia and the US.

“It seems just we think about why a drug might take 15 to go to market…”—

AI Math Breakthroughs

Brockman presents recent math results as evidence that AI can generate new knowledge, while arguing that the larger opportunity is empowering people to solve problems across fields.

Human Insight In Mathematics

Brockman says mathematical discovery should produce insights people can understand, and describes debate-style methods that help humans evaluate difficult AI-generated work.

Biology Data Bottlenecks

Brockman links AI progress to compute, data, and algorithms, arguing that biology could benefit substantially but that scarce data and practical research bottlenecks will shape results.

Drug Discovery Potential

Brockman identifies drug discovery as a major AI opportunity, citing the potential to accelerate research and find treatments while noting that clinical development remains lengthy.

Clinical Trial Expectations

Brockman argues that regulation should adapt if AI can reliably predict drug safety or efficacy better than current processes, while emphasizing the need to validate long-term outcomes.

DOTS Workplace Agent

Brockman describes DOTS as an always-on assistant that tracks tasks and context across work tools, and explains how compute costs shape workplace and consumer AI products.

AI Interfaces And Personalization

Brockman expects AI interfaces to become more conversational and adaptive, with generated visual tools when useful and enterprise controls where organizations require them.

Cybersecurity Defender Window

Brockman says defenders can use frontier AI before powerful cyber capabilities spread widely, finding existing vulnerabilities and addressing long-standing security debt.

Securing Software With AI

Brockman describes OpenAI's process of using new models to probe and fix its systems, and argues AI can make secure coding and software verification more practical.

Security Spending Shift

Brockman argues that security spending may need to exceed software spending as software becomes cheaper, and calls for coordinated work to secure supply chains and infrastructure.

Human Agency And AI

Brockman says OpenAI's mission is to keep humans in control and use technology to improve human lives, even as AI systems surpass people in particular capabilities.