Revolutionizing Cybersecurity and AI-Driven Science with OpenAI President Greg Brockman

Full analysis

Greg Brockman went on No Priors and made a pile of big claims: GPT-6 is live for a billion-plus people, OpenAI's agents cracked nearly 400 century-old math problems, a new always-on agent called "Dots" watches your email and calendar without being asked, and cybersecurity is about to become the biggest software spending category because AI makes code cheap and security scarce. He also threw down a competitive moat claim: nobody else has both a billion consumers and a leading enterprise coding business under one roof.

Here is the frame for a business reader. Almost none of this is a decision you have to make this week. It is a frontier lab telling you where it thinks the puck is going, from the mouth of the person with the most reason to make OpenAI look inevitable. The question worth asking: which of these claims should change what you buy or build, and which is the view from the top of the funnel? Everything here is easy to walk back if you wait. There is no deadline, no price change, no contract cliff in the episode. So the right posture is to separate the verifiable from the promotional.

The Skeptic. Count the unverifiable claims. "Nearly 400 open math problems resolved" with "beautiful insights" that experts are "actively mining": no paper link, no list, no named mathematician in the summary. Brockman says the proofs contain real human-readable ideas, which is exactly the claim you cannot check from a podcast. Same with GPT-6 at "1.2 billion consumers": a usage number, not a capability number. And the cybersecurity pitch arrives wrapped in a $1 billion commitment and a CrowdStrike logo, which is how you sell a category, not how you prove one. The most checkable claim, "security spend should vastly outstrip software spend," is the one with zero evidence behind it. It is a wish about OpenAI's future revenue.

The Researcher. Strip the marketing and one claim actually matters: that AI generated genuinely new math with interpretable proofs, not brute-force answer-checking like the old Four Color Theorem work. If true and published, that is the first clean public evidence of frontier models producing new knowledge a human can learn from. But "if published" is doing everything. DeepMind has walked this road with AlphaProof and AlphaEvolve and published specifics you could inspect. Brockman gave a round number and an adjective. On biology, his "no reason a neural net is worse at biology than language" line is a first-principles argument, not a result. He admits the bottleneck is data, which means the hard part is wet-lab experiments and regulatory filings, the stuff models do not solve by getting smarter.

The Open-Source Advocate. The quiet claim here is the moat: a billion consumers plus a leading coding and enterprise business, "no one else has both." That is an argument for lock-in, and it should make you nervous about building your whole stack on it. On the capabilities that matter to most buyers, the gap to open weights keeps shrinking. For coding and agent work, Qwen, DeepSeek, and the Llama line are close enough that plenty of teams run them in-house for a fraction of the API bill. None of what Brockman described, proactive agents, generated interfaces, security hardening, requires OpenAI specifically. The math and drug-discovery frontier is genuinely theirs for now. The everyday builder's workload is not.

The Compute Pragmatist. Follow the compute. "Dots" gives each user their own always-on cloud computer that monitors Slack, email, and calendar continuously. It runs for you around the clock, and someone pays for those cycles whether or not you prompt it. Expect this to arrive as a new, pricier tier, because persistent agents burn tokens while you sleep. Then read the cybersecurity pitch through the same lens: "security spend should outstrip software spend" is the economic story a company tells when its real problem is that inference is expensive and it needs a high-value category to sell those cycles into. Security budgets are a place to put a lot of compute at a price buyers will tolerate.

The Builder. What could you actually ship Tuesday? The generative UI idea, where the model builds a flight-booking screen or a bike-assembly guide on the fly instead of a developer coding it, is the most grounded thing here, and it is the most dangerous to build on. A UI that regenerates per request is a UI you cannot test, cannot cache reliably, and cannot promise an enterprise buyer looks the same twice. Dots is more interesting as a pattern than a product: an agent with memory across your tools that surfaces what you missed. You can approximate that today with existing models and your own glue. None of this is new raw capability. Brockman said so himself about Dots: "form factor, not capability."

The real disagreements. The Researcher wants the math result to be real because it would matter enormously; the Skeptic notes it is the one claim structured to be unverifiable from where we sit. The Compute Pragmatist and the Skeptic land in the same place from different roads: the "security outstrips software" claim is less a prediction about your budget than a plan for OpenAI's revenue. And the Open-Source Advocate splits the frontier from the floor. OpenAI probably does lead on new-knowledge math and drug discovery. For the coding, agent, and interface work most buyers actually run, the moat is thinner than the moat claim.

What it hinges on: is the 400-problem result published with specifics an outside mathematician can check, or does it stay a round number in interviews? That question decides whether "AI generates new knowledge" is a 2026 fact or a 2027 promise. Before you change any roadmap, wait for the paper, and price Dots as a compute tier, not a feature.

Watch the gap between the publisher thread and the lab's pitch. In Ken's saved reading, Reach CEO Piers North sent OpenAI a cease-and-desist over scraping and is holding up a Google content deal as the template. Brockman talks about AI generating new knowledge; publishers are busy proving how much of the old knowledge still comes from them.

Prediction: OpenAI will not publish, before its next major model release expected by mid-2027, a peer-reviewed or preprint paper that lists the roughly 400 resolved open math problems with proofs and named external mathematicians confirming the insights are novel.

Confidence: Medium. The claim is structured for interviews, not verification.

Why: Brockman gave a round number ("almost 400") and an adjective ("beautiful insights") on a podcast with no paper, no list, and no named mathematician, which is the pattern of a capability pitch rather than a published result. DeepMind, the obvious comparator because it is the only other lab chasing frontier math proofs, backs its claims with inspectable specifics like named competition problems and verifier details; OpenAI did neither here. The opposite outcome, a full annotated list with outside confirmation, is possible but cuts against OpenAI's recent habit of announcing capability milestones through demos and executive interviews rather than checkable artifacts. If the result were clean and ready, the round number and the missing list would be strange.

Revisit by 2027-06-30: We're right if no OpenAI paper or preprint by then lists the specific resolved problems with external mathematicians confirming the proofs contain novel insight. We're wrong if OpenAI publishes such a list with named outside confirmation.