What this covers
In this exclusive interview, Axios co-founder Mike Allen sits down with OpenAI CEO Sam Altman to discuss the rapid approach of superintelligence and why he believes the world, especially Washington, is not yet prepared for the transformation ahead.
Altman details OpenAI's new policy blueprint, the urgent need for societal resilience against cyber and bio threats, and whether the U.S. government should play a more direct role in the development of AI.
Timestamps: 00:00 – Intro: Mike Allen and Sam Altman on "Superintelligence" 00:39 – Why Sam Altman feels a sense of urgency right now 02:17 – OpenAI's responsibility as an "educator" for what's coming 04:14 – What to expect from the next class of AI models 05:21 – Serious threats: Cyber security and Bio-risks 07:47 – The case against nationalizing OpenAI 09:36 – How far are we from AGI and a "new paradigm"? 10:55 – Will AI transform what it means to be human? 12:50 – Is the AI blueprint "Bernie Sanders' fever dream"? 14:50 – What is most doable in the policy blueprint? 16:33 – Personal impact: How leading OpenAI has changed Sam Altman 19:00 – Rapid Fire: The future of AI pricing and intelligence as a utility
Read OpenAI's new report: https://openai.com/index/industrial-policy-for-the-intelligence-age/
Source description (no synthesized summary yet).
Altman argues that AI capabilities are advancing faster than society is prepared for, and both government and business must urgently align on policy, infrastructure, and safety frameworks to ensure superintelligence benefits democratic societies before authoritarian regimes develop it.
- Current AI models are already making legitimate scientific discoveries and doing economically valuable work at scale, with next-generation models representing a 'very significant step forward'
- Superintelligence development requires urgent action on cybersecurity, biosecurity, and resilience rather than waiting for perfect solutions
- Public-private partnership and democratization of AI are necessary because no company or government can handle these risks alone
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The existence of ChatGPT's billion-user base means that subtle design choices made by a few engineers have direct impact on a scale and reach unprecedented in human history—one company has never shaped the communication style and outputs of a billion people before
“I remember a conversation I had with somebody who works on our post training team. The thing that's sort of like the team that decides how ChatGPT is going to feel, to talk to, and she was saying something like, you know, never before in human history have close to a billion people been talking to the same, like virtual brain. And these changes that we make, you know, these like subtle changes to it, does it, you know, talk a little bit more in this style, a little bit more in that style. The the impact of that is just not something that any of us that no single person could have, like, you know, no single person could talk to a billion people every day or whatever.”
Democracy requires managing system chaos and difficulty that goes beyond ideal efficiency, and throughout his lifetime, functional government has been rare but the country has still made great strides despite this imperfection
“To zoom out for a second, Sam, the premise of all of this is a functional government. Where are you going to find that? In my lifetime, I don't think we've ever had a super functional government. And yet we have managed to make great strides as a country. I think part of a democracy is that you have to manage system chaos, and stuff is not always as easy as you would like. I'm always biased, like, too optimistic, but I assume we'll figure it out.”
Close public-private partnership between AI companies and government is essential and currently insufficient; this partnership must be deeper and more extensive than standard government-industry relationships to address the scope of AI transformation
“However, I do think absolutely, the companies developing this and the government have to work extremely closely together. I don't think this works as like a standard or a company in you're the government. There's going to have to be very, very deep partnership here. But that's not happening now.”
Major swaths of knowledge work have already been transformed by current AI models, as demonstrated by the significant difference in what it means to be a coder between early 2025 and early 2026
“I would say already, major swaths of knowledge work have been transformed. The world has, you know, adopted that at different rates. But certainly what it meant to be a coder at the beginning of 2025 versus the beginning of 2026 were very different things.”
Current AI models are capable of helping scientists make small to significant discoveries, with the next generation expected to enable career-defining scientific breakthroughs comparable to major research accomplishments rather than individual Nobel Prize-level work
“Let's say that the current models can help scientists make small discoveries. I really don't want to overstate what the current models are capable of, but you see these amazing things from scientists on, you know, Twitter or other platforms saying, I use GPT 5.4 and it help me do this amazing thing, or that amazing thing. And, I would expect that with the next class of models, you start to see people say, this helped me make the most important discovery of my decade or maybe my career. So that kind of level like, you know, maybe not win a Nobel Prize on its own, but like a significant, career defining discovery”
Government nationalization of AI development is unlikely to succeed in the current global context because the US needs to build superintelligence aligned with democratic values faster than a government bureaucracy could move, and private companies are better positioned to win this race against other nations
“I don't think in the current way the world works, that is likely to be successful. And so I think the biggest case against nationalization would be that we need the US to succeed at building superintelligence in a way that is aligned with the democratic values of the United States before somebody else does. And that probably wouldn't work as a government project. I think that's a sad thing.”
The price per unit of intelligence will continue to fall astronomically fast if sufficient infrastructure is built to avoid supply constraints; if there is not enough infrastructure, prices could spike due to demand exceeding supply, which would be negative and is why massive infrastructure investment is necessary
“The, the price per unit of intelligence, I suspect to continue to will will continue to drop. Incredibly, if we can build enough infrastructure, there's so much demand for this in the world that I can see like a very weird supply constraint. And then prices go up. I think that'd be bad. That's why we should build so much. But assuming we can build enough, I think like the price per unit will continue to fall at the rate it's been falling, which is astronomically fast.”
Building electricity and energy infrastructure faster to keep electricity prices down has strong political support across both parties and represents the most immediately actionable policy idea from the AI blueprint
“Honestly, I think where there's the most energy right now for something tactical is around energy, electricity prices and the desire to figure out a way that we're going to build more energy faster to keep electricity prices down seems to have quite a lot of political support. So that's one thing where I would say like this is pretty standard.”
Frontier systems auditing is a complex policy area where balance is needed between avoiding overregulation and managing risks appropriately; this sits between simple/mainstream ideas (energy) and currently out-of-window ideas (economic transformation)
“And then and then there's like ideas in the middle about like, okay, how much auditing should we have on Frontier Systems? We don't want to overregulate, but we don't want to big risk. So I'd say those are kind of like feel more complex.”
Frontier AI models will command a price premium because users will always demand bigger and smarter models with more capability, and these premiums will persist until the cost to produce frontier models eventually drops to commodity levels like previous-generation models
“I also think that we have continued to see that people will pay a lot more for the frontier of intelligence. So you'll have like the price of last years of intelligence. Fine, fine, fine, fine, and the price of the frontier. And people always want a bigger, smarter model thinking more that'll keep going up, because one of the guys who's a partner in a startup said, I'm legitimately worried a change in AI pricing could turn our business upside down. So far, pricing has only gone in one direction and again, without something weird happening around us not building enough supply, I'd expect that to continue to go forward. But I also think the world clearly wants bigger, smarter models, and those will cost a premium until their cost comes down to some of them.”
People building AI should be high-integrity, trustworthy individuals who feel the gravity and responsibility of the work; additionally, no single person should make decisions unilaterally affecting all of society, so democratization and public understanding of AI rules is essential
“On the trust point, it's incredibly important that people building AI are high integrity, trustworthy people. And I think almost everybody involved in our industry feels the gravity of what we're doing. And so we all take that responsibility very seriously. We feel that weight every day. We also think it's very important that no one person is making the decisions by themselves that are going to impact all of us. I don't think we should have to trust a single person to get every decision right. We really believe in the democratization of AI. We really believe in putting this tool in the hands of people and letting people understand it. Figure out what rules society collectively wants to put around it, and most importantly, how we're going to integrate it into our lives and have all these incredible benefits that we think are possible.”
The current frontier models are all in the hands of relatively responsible companies, allowing mitigation of biosecurity risks through company-level alignment and safety mechanisms
“When the models, the frontier models are all sort of in the hands of pretty responsible companies, I think we can mitigate those by the companies aligning the models and having good classifiers and good safety stocks.”
The main near-term safety risks being tracked are cybersecurity threats and biological threats; in the next year, there will likely be significant cybersecurity threats requiring mitigation, and biological misuse becomes theoretically possible soon as models become capable at advanced biology
“The main areas we currently track in our preparedness framework are cybersecurity, bio, and actually any issue of time. Let me stop at those two. I suspect in the next year we will see significant threats we have to mitigate, from cyber. And these models are already quite capable and will get much more capable. And then on bio, this is something we've been talking about a lot. The models are clearly going to get very good at helping people do biology at an advanced level. Wonderful things are going to happen there. Will see a bunch of diseases get cured. Someone is going to try to misuse those”
It is possible that AI technology follows the expected development path but society takes much longer than anticipated to adapt, resulting in delayed impact compared to technical capability timelines
“We may be wrong. It is possible that even if the technology stays on the path that we expected to stay on, society takes much longer to adapt to some of these changes. Then, than we would expect.”
When GPT-4 was finished but not released, the knowledge that AI capabilities had reached a transformative level, combined with the weight of deciding how to release it and the impact of that choice, was the strongest Altman has felt the personal significance and change-making power of his role
“That the strongest I ever felt it was when we were, we had finished training GPT four. We didn't release it for many months after we finished it, but we knew we had this incredible thing. We knew that the world was never quite going to be the same, and we knew that the way we released it was hopefully going to nudge things in a better but possibly a worse direction. And that time where this relatively small number of people in the world knew this thing and we couldn't talk about it yet, and we knew that it was going to have this gigantic impact. I think there's no way to live through that, not have it change you for good and bad. Mostly for good, but in all sorts of ways.”
AI will not transform the fundamental aspects of what makes life meaningful for humans—people will still value deep personal connections, time in nature, and activities that make them happy, even as the economic structure and organization of society changes significantly
“First of all, not literally every. I think there will be many things, and I think this is wonderful. Where we only care about other people, we will spend more time with other people. People will remain the most important part of our lives. You know, we'll have these incredibly smart machines doing stuff around us. And the most fundamental part of what it means to be human, that's not going to transform at all. And I think this is really important. Like the society may change in a lot of ways what it means to be a, you know, what it means to be a person, what it means, some of the civilian life, how we choose to spend our time, what we really want. We'll have more flexibility and more ability to achieve. But, you know, for hundreds, thousands, tens of thousands of years, we've been driven by a lot of the same things. We want more of those. We want higher quality connections with people. We want more time to, you know, spend in nature, do what makes us happy.”
It is a 'sad thing' that government-led development of AGI would likely not be successful in the current era, as it would represent a departure from historical patterns where transformative large-scale projects were government-led
“I think that's a sad thing.”
The goal of OpenAI's policy agenda is to start a broad conversation and debate about AI governance, not to impose a single solution; they expect many better solutions will come from wider participation and debate than their initial ideas
“We want to put these things into the conversation. Some will be good, some will be bad. But as you said, we do feel a sense of urgency. And we want to see the debate of these issues, really start to happen with seriousness. Again, there probably be many more solutions, better solutions than we can think of ourselves and the broader world.”
OpenAI has a responsibility to educate the public about what they see coming from their research pipeline, and while they have sometimes been wrong and will be wrong in the future, this educational role is part of their contribution to society
“Part of what we try to do throughout the history of open AI is just talk about, with appropriate uncertainty, what we see coming. And we have not always been right, but sometimes, pretty often, I think we have been and and I think we have a responsibility as to play a sort of role as an educator on what we see coming down the research pipeline. For a long time, we've said things that at the time we said them originally were considered like, you know, some more polite version of crazy. And again, sometimes we've been wrong and we'll be wrong about some things in the future, too.”
Both Democratic and Republican politicians and policymakers have shown receptiveness to the idea that AI requires new economic and policy approaches, suggesting broad recognition that the current economic model may need adjustment
“We've talked to Democrats about some of these ideas. We talked to Republicans about some of these ideas. Again, as I said earlier, I'm sure they aren't all good. And the goal here is really to start a conversation. That said, the receptiveness from all directions to the fact that we are going to have to try some different things, and the sooner we can start talking to them, the better, has surprised me. On the upside”
Some of OpenAI's early predictions that seemed crazy at the time have turned out to be important and accurate, suggesting that their current urgent claims about superintelligence should be taken seriously even if they seem implausible now
“For a long time, we've said things that at the time we said them originally were considered like, you know, some more polite version of crazy. And again, sometimes we've been wrong and we'll be wrong about some things in the future, too. But part of our job, and I think part of how we can contribute to society is say, here's what we see coming. We may be wrong. We may be right if we are right. Here are the kinds of ideas that we think might work”
AI pricing changes could fundamentally disrupt startup business models if pricing direction reverses or becomes unpredictable, but Altman expects pricing to continue falling in one direction absent major infrastructure supply shocks.
“one of the guys who's a partner in a startup said, I'm legitimately worried a change in AI pricing could turn our business upside down. So far, pricing has only gone in one direction and again, without something weird happening around us not building enough supply, I'd expect that to continue to go forward.”
Altman lives with significant stress and anxiety due to his role in AI development and no longer has a normal life; he acknowledges the burden is real even if it's uncertain whether he ever had a truly 'normal' life
“I live with a lot of stress, anxiety, weights, whatever. And I don't think I have a very normal life anymore. Did you have a normal life for, relatively to know for sure?”
Altman is biased toward optimism and assumes democracy will figure out how to adapt to AI transformation, despite current dysfunction.
“I'm always biased, like, too optimistic, but I assume we'll figure it out.”
When in the immediate tactical work of running AI development, Altman often doesn't feel like he's doing something historically significant; he's just solving problems.
“when you're in the grind, you don't often feel like I'm doing this like, you know, big, historic, weighty thing. You're just kind of like trying to solve whatever the tactical problem in front of you is”
Large shifts to how society functions economically, such as changes to tax systems or broader economic structures, are still outside the Overton window but near its edges, requiring longer-term political development.
“the idea is we're talking about a little bit earlier, like the big sort of shifts to how you make a change to the tax system or the economy. In a larger sense, I'd say those are still now in the Overton window, but near the edges.”
The phrasing 'intelligence is a utility like electricity' was meant to describe the functional and pricing model rather than a literal pricing mechanism; the electricity analogy illustrates how users will experience AI as transparent infrastructure they plug services into
“I didn't necessarily mean that about pricing, but I'll get to that to what I meant. Like, you know, you pay an electricity bill to your house and then you plug in whatever you want, like you have your computer, your oven, you know, like my kids toys, whatever. All use this electricity.”
Privacy and AI is an area where something substantial will happen with bipartisan support and without significant controversy.
“There's other ideas like, you know, privacy and AI that also I think something will happen and it's pretty non-controversial and there's like quite, quite a bit of support”
Scientists are already posting on social media about using GPT-4 for significant research contributions, providing observable evidence that current models are having real scientific impact.
“you see these amazing things from scientists on, you know, Twitter or other platforms saying, I use GPT 5.4 and it help me do this amazing thing, or that amazing thing”
OpenAI does not currently overstate the capabilities of existing models, instead being cautious about claims.
“I really don't want to overstate what the current models are capable of”