“Death becoming a choice”: what Greg Brockman said about OpenAI's plans

OpenAI co-founder and president Greg Brockman sat down with The Wall Street Journal's Keach Hagey on stage at WSJ Tech. The recording was published by the WSJ Events channel on 4 November 2026. The conversation took place the day after the US midterm elections and covered almost everything the company is dealing with right now: local opposition to data centres, cybersecurity, revenue, going public, advertising, open models and political money.

The loudest line came near the end. Asked how superintelligent AI would change life, Brockman said we could even be thinking about death becoming a choice, and that significant strides in biology, health and curing disease would be visible within the next 12 to 18 months.

In brief

• Medicine. Brockman expects real results in biology and disease treatment within 12–18 months; the full impact, he says, will take longer. Death as a choice is presented as a possible goal.
• Cybersecurity. With roughly a million dollars of compute, OpenAI's models found about a thousand vulnerabilities in Linux and produced patches for them. Typically, Brockman says, Linux sees around twenty such findings a year.
• Money. In the third quarter, annualised revenue grew 70% quarter on quarter, and the enterprise side doubled. Enterprise customers now account for the majority of revenue, up from about 30% a year ago.
• New customers. A partnership with JPMorgan Chase was announced: the bank will use Codex and OpenAI's cyber models.
• IPO. Asked whether OpenAI would be ready to go public next year, Brockman said: “I think so.”
• Ads. ChatGPT's advertising business went from zero at the start of the year to a billion dollars in annualised revenue.
• Politics. Brockman and his wife pledged $50 million to the Leading the Future super PAC, paid the first $25 million and decided not to pay the second tranche.

Death as a choice

The question was about how AI might help not only people but animals too — something Brockman and his wife have spoken about publicly. He answered through biology: it is shared by all species, and so are the benefits of understanding it. Then came the key line: one can even think about death becoming a choice, and all of this, he said, is possible and doable.

Brockman himself acknowledged that curing diseases has become a trope in conversations about AI. But he insists that this time there will be real delivery, and soon. His logic is simple: the models are already solving previously unsolved problems, so they clearly need to be pointed at biology and the life sciences.

He also offered a five-year forecast. The question “do you use AI?” will sound as odd as “do you have electricity?” People will spend less time typing at keyboards and more time interacting with each other. There will not be less work, though: he points to software engineers who, in his view, have become far more productive and yet work harder than ever, because there is simply so much opportunity.

A million dollars on one problem

Brockman described a new way of working he calls agent fleets: many AI agents and around a million dollars of compute pointed at a single hard problem. According to him, this is how OpenAI solved the Navier–Stokes problem, one of the famous open problems in mathematics. The interview gives no details, and the claim needs separate verification by mathematicians.

The same approach was applied to cybersecurity: about a thousand Linux vulnerabilities with ready patches in one pass. Hence, Brockman says, the new business: not speeding up familiar workflows by a few per cent, but solving problems that would otherwise remain unsolvable. The question he puts to customers: where else can you point a million dollars of compute?

The defender's window

Asked whether the industry had traded enterprise productivity for an endless cyber arms race, Brockman answered with history. The internet was not built for security, encryption was a retrofit, and forty years of technical debt have piled up. AI, he says, is simply shining a light into the cracks everyone has created.

The way out, as he sees it, is to put the best AI in the hands of defenders before it reaches attackers. OpenAI calls this gap the defender's window. Ideally, it leads to a world where breaches are unthinkable: code with formal security guarantees, written by models that do it better than people.

The most controversial idea is an always-on system that scans for vulnerable servers, and even a right for the government to apply patches to other people's machines itself. Brockman admitted the idea is controversial: such an intervention could take a service down. But he cited a precedent: over roughly the past five years the FBI has several times obtained court orders to clear malicious botnet code from hundreds of thousands of private computers. He said OpenAI has started floating the idea with the administration and is warning governments in several countries that the next 6–18 months will be a period of mass vulnerability discovery. A quarterly patch cycle will no longer do: it needs to come down to a month, a week, and ultimately minutes.

Data centres and the elections

The conversation opened with politics: many of the Democrats who won the elections support moratoriums on data centres. Brockman admitted that the unpopularity of data centres surprised everyone and that American politics has seen nothing like it. The company's mistake, as he sees it, was trying to build infrastructure in Texas and Georgia while staying in San Francisco. You cannot win people over with tweets: teams need to be on the ground and stay, not show up for a week. Among the local benefits he mentioned were free lunches for students in Georgia sponsored by OpenAI.

Revenue, IPO and ads

Brockman admitted the first two quarters of the year were rough: in coding, he says, OpenAI was not even in the conversation, although it had Codex. The turning point came in the third quarter, after the release of GPT-5.6 Sol and ChatGPT Works. He stresses that this was the result of earlier investments, not one lucky quarter.

The main shift is in the structure of the business. A year ago OpenAI was seen as a consumer company; now enterprise customers bring in most of the revenue. Brockman called the JPMorgan Chase deal iconic: a highly regulated bank for which trust is everything.

The interviewer raised the awkward points: Anthropic is going public before OpenAI, and OpenAI, according to the WSJ, raised less than half of what it targeted in its latest round. Brockman replied that the company will go public when it feels the time is right, and that he has never felt more confident about the business. On efficiency — Anthropic reported that its revenue per dollar of compute expense jumped 5.2 times, from 40 cents to over $2 — he gave no comparable figure for OpenAI. Instead, he talked about price cuts: the Luna model became 80% cheaper, after which its usage grew more than tenfold. For the strongest models, he says, customers are almost price-insensitive, and margins there can be tuned almost arbitrarily.

Brockman said ChatGPT's ads business has been hockey-sticking: a billion dollars in annualised revenue by around the summer, from zero at the start of the year. He believes no one has done this before.

An always-on assistant

They also discussed products that run all the time, such as the Dot assistant. Brockman acknowledged that keeping a model on all the time is expensive, but the company's philosophy is to build the peak experience first, even if it is out of reach for most people, and then make it cheaper. His example: the day before, Dot flagged a very well-crafted phishing email in his inbox because of a domain mismatch. Brockman says he would never have noticed it himself.

Open weights and political money

Brockman does not call for banning open-weight models. He considers it important to democratise AI and to avoid a full concentration of power, which he says is part of OpenAI's DNA. At the same time, in his view frontier models stay ahead of open ones, and that buys time to try out dangerous capabilities in a restricted form.

On the pro-AI super PAC Leading the Future, Brockman said that he and his wife see safety and innovation as inseparable, and that, rightly or wrongly, the PAC came to be perceived as opposed to safety. So they did not pay the second $25 million. Asked whether he regrets the first, he gave no direct answer: he is looking to the future.

When asked about many safety researchers putting the chance of AI killing everyone within a decade at over 10%, Brockman did not give his own number either. It is hard, he said, to multiply numbers together and reach confident conclusions; he joined OpenAI so that AI, which is coming anyway, would play out at least slightly better with him than without him.

Editor's view

The line about death as a choice is not a slip but a deliberate image, and it says a lot about how OpenAI now speaks to the world. A company that needs investors' money, regulators' goodwill and residents' consent to data centres next door explains its mission with the strongest promise imaginable — victory over death. And it attaches a deadline: 12–18 months to noticeable results in medicine.

These words deserve a sober reading:

• AI does not set the timelines in biology. Even if a model proposes a drug tomorrow, testing it in humans takes years. Brockman himself says the full impact will take time, but it is the 18 months that make the headlines.
• Big scientific claims without details. A solved Navier–Stokes problem and a thousand Linux vulnerabilities are claims made by the company from a stage. Until mathematicians and kernel developers check them, they are announcements, not facts.
• The awkward questions got no numbers. Compute efficiency, the funding shortfall, a personal estimate of catastrophic risk — each answer drifted into generalities. The most concrete figures in the interview are revenue growth rates.
• A government with the right to patch. The idea that authorities could fix vulnerable servers themselves sounds sensible against botnets, but it means default state access to private machines. It should be discussed not only with the US administration, but with everyone whose servers would fall under such oversight.

There is also an admission in the interview that is rare for an executive at this level: trust cannot be bought with tweets from San Francisco. Protests against data centres show that society is starting to hold AI companies to account not for promises, but for consequences close to home. The promise of immortality sounds beautiful, but people are asking about electricity, water and jobs today.

One last point. If death really does become a choice, the main question will be social rather than technological: who gets that choice. Brockman speaks of democratisation and avoiding concentration of power, yet he himself describes a world where the strongest models are expensive, customers pay for them almost regardless of price, and access to the most dangerous capabilities is decided by a single company. There is no bridge between these two pictures yet.

Source: Watch Live: OpenAI Co-Founder Greg Brockman at WSJ Tech. WSJ Events, YouTube, 4 November 2026 (Greg Brockman in conversation with The Wall Street Journal's Keach Hagey at WSJ Tech). Watch on YouTube.