Coding was the first business OpenAI proved it could sell. Greg Brockman, the company’s president, and chief financial officer Sarah Friar now say the next one could be larger: cybersecurity.
The two made the case in an interview, describing security as a bigger commercial opportunity than software development. People who have spoken with Brockman say the subject has become something close to an obsession, and that he has been pressing internally on where the company should place its bets.
The logic behind the claim is not complicated. Software development is a market of a few tens of millions of professionals, and the tools that serve them are priced per seat. Security is a market of every organization with a network, a compliance obligation and an insurance policy, and the spending is already enormous relative to headcount because it buys coverage rather than labor.
Security work also fits the shape of what large models do well. The job is mostly reading: logs, alerts, configuration files, code, threat reports, and the records of what happened the last time something similar appeared. A system that can hold a million tokens of that material in one pass, correlate it, and write a finding in the format an auditor expects is doing something an analyst does slowly and inconsistently.
The harder half is trust. A model that triages alerts and gets one wrong in a hundred is a nuisance. A model that closes a ticket and gets one wrong in a hundred is an incident with a name. Security teams buy tools that fail in visible, bounded ways, and they are slow to hand judgment to anything that cannot explain itself to a regulator after the fact.
That is where the commercial argument meets the product problem. Selling coding assistance to developers is a bottoms-up motion: a team tries it, likes it, and the purchase follows. Selling security is a top-down motion that runs through a chief information security officer, a procurement process, a proof of concept against live traffic, and often a legal review of where the data goes.
Brockman’s interest also reflects a shift in what OpenAI sells. The company began as a research lab with an API, became a consumer subscription business, and has spent the past two years building the machinery for enterprise contracts: sales teams, compliance documentation, private deployments, and finance functions that can forecast usage. Friar, a former finance chief at Square and chief executive of Nextdoor, was hired into a company that needed exactly that.
The security market has an unusual structure that helps a new entrant and hurts it at the same time. Buyers already run a stack of vendors, each with a console, a data pipeline and a renewal date. Replacing one is a project; adding a layer above them is easier, and an AI system that reads the output of the existing tools rather than replacing them can enter without a rip-out.
The incumbents are not standing still. The largest security vendors have been buying artificial-intelligence capability and building agents that investigate alerts on their own, and they own the telemetry that any model needs to be useful. An outside lab arrives with better models and no data, which is the reverse of the position those vendors hold.
Regulation cuts in an unexpected direction. Disclosure rules in the United States and Europe have pushed companies to report incidents faster and describe them more precisely, which creates demand for tools that can assemble that record under deadline. The same rules make an automated error more expensive, because a wrong filing is itself a violation.
There is a pricing question underneath all of this. Coding tools are sold per developer per month, a model that buyers understand. Security could be sold per endpoint, per alert, per investigation, or against the value of the losses avoided, and each choice produces a different revenue curve. The most attractive version, from a vendor’s view, is priced per outcome, which requires the vendor to accept some of the risk.
OpenAI has not published a security product, a customer count or a revenue target. What it has is a president saying publicly that the market is bigger than the one that made the company’s name, and a finance chief who has to make the numbers work. Statements of that kind usually precede a build rather than follow one.
The comparison to coding is instructive for another reason. Coding became a real business because the output could be checked: code runs or it does not. Security findings are judged by whether something bad happened later, which is a much slower and noisier signal, and one that customers attribute to luck as often as to tooling.
For the buyers who will be pitched this product over the next year, the useful question is not whether a model can find a vulnerability. It is whether the model can be handed a queue, left alone overnight, and trusted not to make the queue worse. That is a narrower claim than the one Brockman is making, and it is the one that will decide whether security turns out to be the bigger business or merely the harder one.


