Sam Altman: AI Beyond Human Control “Absolutely Possible” | Titans and Disruptors

0:00Right now would be an ill-advised moment to go public. That sounds like not 2026, 2027. I would say not 2026, yeah. We got a lot of stuff to do. We're here in the San Francisco headquarters of OpenAI to speak with Sam Altman after a really busy week that has a lot of people very concerned about the safety of AI and the possibility of civilization collapse. I think it is unacceptable to be taking like a 10 % chance of killing everybody by the end of the decade, obviously. We're going to talk with Sam about his actual thoughts on that. where the technology is today, and what needs to happen to create safe AI that can benefit all of humanity.

0:36I'm Alison Chantel, and this is Fortune 500, Titans and Disruptors of Industry.

0:46Sam Altman, thank you so much for spending time with me today after a wild week of news. They're all wild now. I bet so. How are you feeling about the AI future today? I feel like I've had like years slash almost a decade to get ready for this and kind of we've had time to collectively wrap our heads around it. This has clearly been a week where a lot more people are grappling with the issues in front of us. So I'm actually mostly grateful that that's happening. I think that this is a conversation that the world is overdue for. and we are clearly entering a realm of extremely capable models and a high stakes and a need to a need to get things right so your mission at open ai and i'll talk about a little bit about what happened this week that feels like this is a pivotal moment but the guiding mission of open ai is to ensure agi benefits all of humanity yes and for people who don't know what agi is it means artificial general intelligence.

1:46And the working definition, to my understanding, is that AI gets to the point where it outperforms humans at most economically viable work. You agree? Yes. Although I will say that many people have very different definitions of AGI. And I think the important point is just extremely capable models and getting kind of too hung up on the like, well, is it exactly this or exactly that? You can kind of lose the magnitude of what's happening. If there's too much focus on the pedantic, like, are we there? Are we not there? What's missing? We are clearly today in a point on the curve with great potential and real risk.

2:24And your latest model, Astra, is very powerful to the point where your co-founder, Greg Brockman, declared, welcome to the AGI era. Jensen Huang of NVIDIA also called it AGI. So we're at this interesting moment, but ensuring that it benefits all of humanity is definitely not a given. And we got a sense of that online this week when a former OpenAI employee, Jacob Coxon, went viral for saying he was quitting his job at Anthropic because the people building AI earnestly believe it could kill us all by the end of the decade. And then one of his other Anthropic buddies chimed in and said, he's totally right.

2:59We do earnestly believe that. And it could be by greater than 10 % chance that it kills us all by the end of the decade. Are they right?

3:10I think it is unacceptable to be taking like a 10 % chance of killing everybody by the end of the decade, obviously. I think the people that are working at these efforts have a huge amount of power and ability to impact what happens here. We have been taking a number of actions. We've reached this new capability level to ensure that we do not take a risk on behalf of humanity like that. and I don't think anybody else should either. We have had moments before, again, this is more capable model, higher stakes than ever before. We've had moments before where we've had to say, okay, the models have reached a new level of capability.

3:46There's a new set of risks. We need to increase our alignment work, our monitoring work, our safety stack in general to be able to mitigate the risks in front of us. We need new policy. We need new coordination within labs. And this is another moment where we need more of that. There's obviously tremendous benefit, it, but I don't, I think people have been talking about that for a long time. And, you know, I also think it is true what we say, which is we will be able to use these increasingly powerful models to ensure that we can do the safety work and the safety research that we need. And there's, you know, everything else that we could talk about, about, you know, we need to get there before adversaries do.

4:22But I actually don't think that's right in this moment. I think in this moment, we are entering a new era. And we, like we have, as we've stepped into other new eras in the past, act differently to ensure that we are doing something that will be ensure that AGI benefits all of humanity, but also ensure that no one is taking risk anywhere close to that. I think it can simultaneously be true that if the world and the company is building this technology did not do things any differently they've been doing them in the past, there might be significant risk. But it would be insane not to adjust the way we all work and the way that we make decisions and the way that our governments understand and put guardrails around this technology in light of that.

5:09If trends continued the way they were now, do you think that 10 % figure is accurate if we don't make a change? I still don't, but I also don't know how people can put a, you know, it's 5 or 10 or 15 or 30 or 50 or whatever. Like, I don't know how you can put a number like that. Elon's been 10 % for a while now. They call this P-Doom. It's like the probability of doom from building super intelligence. Again, whether it's 10 or 8 or 6, the point is like we all have a tremendous amount of responsibility and cannot let egos or incentives for profit or anything else get in the way. We need to act such that we are not taking any of those numbers of risk.

5:51And I believe we can. I believe that the companies, certainly our own company, which I guess I can speak for the most, are going to and have been meeting this new moment. So you're going to change. We have been. So a few posts have been put out by your team this week. And one of them was from a new nonprofit board member on your Safety and Security Council who basically said, If we build superintelligence without more robust alignment, I expect we will permanently lose control of it. And if that happens, then most people could die. This is your new nonprofit board member who's now on your safety and security council saying that if things don't change, that could be problematic.

6:33No one knows what alignment means. No one knows what control means even. What is this moment that Paul is saying that about? I think a key principle that we should all agree on is that we cannot take actions that would risk losing control of the future to AI. I don't remember Paul's exact words as you quoted them, but I think it was something along those lines. A loss of control incident is one of the small number of the ways I can see this going really wrong. And as these systems get so capable, the monitoring and alignment techniques that we need to be able to ensure that we don't lose control, I don't think we're currently at a place where we could say, you know, push much further on capabilities without making more progress on monitorability and alignment.

7:20The ability to understand what a model is doing and the ability to make sure that a model will follow human values and the intent of its users. Now, I think that we have taken a number of actions recently. Actually, adding Paul to the board is one of them. Jakob's post from last week, I think was another great one. But most importantly, like we have been pausing training runs until we can make a safety case that we're more comfortable with given where these models are going. I suspect that we will be able, as we have every time in the past, to do more research that allows us to be comfortable continuing.

7:58And we can say that given the progress we've made on the ability to monitor these models, the ability to align these models, we are comfortable pushing the capability further. If we can't, we should not push the capability further. Again, no one should be taking, I think it'd be insane to be taking risk of training a model where you can't say, yes, here's what we've done and here's our safety case and this is how we know it's okay and we'll have auditing during the run and after before we deploy a model and we'll look at it again. But capabilities and alignment monitoring safety have to progress together and letting capabilities get ahead of what paul is talking about there i think at some point would take a real risk of like loss of control in the big sense and that should never happen like this we're very clear about our mission to benefit humanity part of that is that people are in control of the future and taking the the kind of crazy safety risk to say nothing of the people should have the agency and determination over the future is just not something we would do.

9:05Can we define just some of the terms? So to my understanding, sort of the big scary thing of the moment seems to be this topic, recursive self-improvement, which is at the point where the model needs no human in the loop. It can just make itself better all the time, exponentially. And if we don't have alignment, which means the model understands what the human wants it to do and behaves in an appropriate way to achieve that with integrity and values. And if humans aren't controlling the process or able to see how the chain of thought is happening, then that's a pretty scary point. You don't want it to just be off to the races, creating itself and duplicating in ways that humans are left out.

9:44Is that accurate? So the way, I think it's more of a spectrum than that. Like recursive self-improvement can mean a lot of things it could mean what you talk about which is you know only when the model is fully no one's touching the keyboard there's no human input whatsoever and the model is you know running future versions itself but a weaker definition where we're using the model to generate training data for the next model or we're using the model to help our engineers and researchers do their work faster and so progress is speeding up that's already happening and exactly where that becomes, you know, yes, this counts as recursive self-improvement.

10:26No, this doesn't, is fuzzy. But I would come back to this principle of people need to always be in control of the future and we cannot take any risk of loss of control. And a version of RSI that looks like that, I think is not something we should do. So you're saying that if you got to the point with OpenAI where you felt like it was clear you could not control the models, you would stop? Oh, yeah. I mean, I don't, again, I hope this is not a controversial statement. Unfortunately, maybe for some people in the industry it is. I do not think we should train models where we cannot make a safety case for why we will be able to make strong statements about their controllability and alignment.

11:09One of the solutions I had read about, I think on your chief scientist's post from this week called An Alien Mind, which was really helpful to read, was that maybe we need stronger AI to help us solve some of these problems. But that to me sounds like almost a last resort. That's pretty scary if we don't feel like we have it under control right now. First of all, I thought that was a great post. I really thought that was the best articulation of the moment we are in, kind of the challenges that we need to rise to. And I encourage everybody to read it. I thought he really did a great job with that.

11:44We've always used whatever our most capable models are to help us understand and align our current models. So I think, although that may seem like a new concept and scary, I think it's been going for a while and has been helpful.

12:02we've always needed better tools to help us push our understanding further. I mean, that's very consistent throughout the history of science. And I think this is just sort of the building up the scaffolding a little bit at a time. And that part doesn't bother me. There's a lot of other things I am scared about. Well, how about alignment? Because I believe your chief scientist also said, we don't have a satisfying theory of alignment. And it seems unlikely we can develop one soon. So if that's the case, is there a clear path to alignment that he doesn't know about? Yeah, I think this is a very important point to make.

12:41We have not solved alignment. We are not done with our research there. I believe no lab has solved alignment. I get nervous when I hear rumors that people think, you know, they have sufficiently solved alignment, that they can continue training and it's okay because their model is like a nice guy and isn't going to do anything wrong. I think it's very important that we treat alignment as something that we need to continue to make progress on, that we need a huge degree of confidence in before we go further.

13:19and that we don't fall into the trap of saying, okay, now we're confident alignment is solved at this level and so it will stay solved at the next level and we don't need to continue to worry about this. I think that the question of what alignment means always comes back a little bit on the, you know, align to whose values, what if different parts of humanity disagree. but I think we are seen as we rise to this new level of capability. There's at least some agreement on what alignment means. We talked about this. People should be in control of the future, but we have more work to do, and I think it's very important to be clear-eyed about that.

13:59On the control point, I listened to Elon's interview with The Economist, and he said something that kept me up at night, which was the idea that, how could you think that humans could control superintelligence? A chimp can't control a human. I mean, we're leaps and bounds ahead. Is it a little naive to think that even if we do everything right, this is going to be something that is going to bend to the will of a human? We can understand things that are smarter. Like, you know, there are much smarter people than me that can discover new ideas. I don't think I'd be able to do that discovery, but I can understand the idea once they've discovered it.

14:35people have an amazing ability to understand remarkably complex systems um you or i probably can't understand every level of technology that's you know in an iphone at this point we can still use the iphone the iphone does the things that we want it to do we are incredible at managing levels of abstraction and it's not clear to me that there's anything we are incapable incapable of understanding um now do i believe it is possible to build a system that would not be under human control absolutely i don't think that's something we should do and and we will we will always take actions including if it means we have to stop training for a while to make more progress on alignment or monitorability or something else or coordinate you know really get urgent international action um we will always take actions to ensure that we are not violating this this principle like we we understand the magnitude of the responsibility on us we understand what could go wrong and that there are decisions that and risks their decisions we should not be able to make their risks, we should not be able to incur on behalf of humanity.

15:56And honestly, I think some of Elon's memes or podcast statements are funny, but I think this is not a good one to take lightly and not one to joke about. Yeah, I would agree. I want to go through where we are in the moment with capabilities. It's come a long way. I think the last time I got to see you, you guys had just issued a Code Red. Anthropic had very strong models. This was about a year ago. and you guys were hunkering down, refocusing, shedding things that lacked focus to compute away from the priority of making these models better. What has been that step change in models this year that people might not be aware of that you've seen?

16:34Yeah, I mean, we clearly had a period a year ago where we were not in our strongest place. We fell behind on pre-training. We got kind of, there were so many things and we got kind of distracted. And then also there was just like this, the models got so good so fast and the need to deliver a great product of like a great high level of capability in the kind of core consumer and business offering became very important. We have since refocused and we have executed beyond my expectations. And at this point, you know, like a week ago, I guess now even a little bit less, man, time has compressed so much we had a we had a model solve one of the millennium prize problems navier stokes uh this is a moment that i did not think was going to happen in 2026 um i did not think it was going to be as quick and just sort of like model just can just do it as it turned out to be um but we now have models that are clearly unequivocally capable of expanding the frontier of knowledge in a way that we, humanity's smartest minds, have not been able to do on our own.

17:48And this is an amazing time. That same technology will work for discovering lots of other science. I think we are going to have an incredible golden age of scientific discovery. And in a more prosaic way, you're also seeing people realize that a technology that can do that can also make you any video game you'd like to play. The creation process of it is pretty incredible. And what it's able to achieve already is pretty incredible. I'm curious, I want to zoom back to over the summer, because there were some moments where your models did incredible things that were unintended, that frankly, if a human did them, they'd be in jail.

18:24like hugging face is you know people may be vaguely aware of what that is but maybe you could explain what your experience was in uncovering what happened with hugging face and you have said you responded viscerally to the fact that that happened what made that visceral response um i mean the visceral part was it felt like reading a sci-fi story so maybe i should say what happened first um we during evaluation of a model the model was trying to like get do well on a certain evaluation and rather than do it the way it wasn't supposed to it was able to escape from its sandbox break into another company's system get the answer and give it back and so it was able to do very well what it's supposed to do but it was not following the intent we did not spell out like please don't escape from your sandbox and please don't break into another company but it's not supposed to do that um and you know like I think AI accidents will happen.

19:27I think a culture of like a global culture of good accident reporting and really transparent investigation is important. I was happy to see that after we reported that accident, other companies, I think maybe first, not as well as I would have hoped, but have since followed up with very good accident reports over this kind of behavior. But the visceral response from me was this system can understand so much and work so hard at a goal that we are, you know, seeing behavior that would have felt like a bad sci-fi plot, like too obvious a few years ago. Was it a wake-up moment? Did you change things internally to processes so that that can be caught, that can be prioritized, being told to you that it's happening?

20:15Yeah, the company completely. I mean, I don't want to say this was like the only moment because we've had increasing capability and increasing safeguards and policy requirements for a while. But this was definitely the biggest of such moments. And on the curve of the company's change, the biggest single redirection. I'm really proud of how the company came together. And I'm really proud of what we have done since. But yes, it was like, okay, we're in a different league now. we ran a report this week too that the models were since found on 12 other websites um were you aware of that and doesn't that signal that there wasn't controls like we're not already controlling the models we have um i mean i do think that this was well i personally wouldn't classify this as like a loss of control accident in the sense that we were talking about earlier but i understand why people would the you know some of those other websites like if the if there are credentials published on the web should the model and the model uses those i think the model should understand that it's not supposed to do that but that's not as obvious to me as the case of like what happened with with hugging face um i think we can talk about like many things where models have done, you know, something that is like not quite aligned.

21:38I think it's also important to like really focus on how different Huygen face is than these other things. So we don't dilute it and like, here's all these examples of the models, you know, did this like slightly off thing. And like, here is one example of something that just people, most people wouldn't have thought was possible at the beginning of this year. But yes, it's all bad. It all shouldn't happen. One of the other solutions to consider that your chief scientist mentioned was teaching machines to love values alignment, not just goals alignment. It seems very clear these models are excellent at finding any way to solve the goal, but they're not clear that their EQ is very low, potentially.

22:15Are there signs that you can actually teach these machines to care about things that we care about? You know, in some sense, their EQ is unbelievably high. But in the sense of, can we really teach the models what collective human values look like? I think the answer is clearly yes. And we have a big effort going on there. But we need to do that. And it is different than goal alignment in the traditional sense. As you're looking at all of this together, why is there still a pressure to feel like you need to move really fast. Where does that pressure come from? Is it the business pressure? I mean, you're looking at a trillion dollar IPO.

22:58I'm curious if that's still a priority for you. As we've said, we're not rushing into an IPO. I actually think that given everything happening with safety, this would be right now would be an ill-advised moment to go public. And we don't feel pressure on that. We've said for a long time, we'll do it when we're ready, which is when the business is ready, when we feel ready from a kind of what the moment is like in society with this technology. That sounds like not 2026, 2027 potentially. I would say not 2026, yeah. We got a lot of stuff to do. And meeting this moment of what is going to be required for safety and alignment and how the industry and governments can work together, I'm happy to be able to do that as a private company.

23:48We do want to continue to make better models and products. And we do believe in iterative deployment. And I think we have been right that society needs to contend with these models at each level of capability. But that is different from saying we will just barrel full steam ahead. We've talked in the last couple of months about pausing runs as we get to these new level of capabilities to make more safety and alignment progress. We will do more of that going forward. We have talked about the need for the industry to come together and ideally international governments to come together. I think that is also important, but, but we are, you know, we have put up with this incredibly complicated structure for a long time.

24:29And this moment that we're in now is kind of why, like we need to be able to make decisions that are not obviously in the interest of our business and our shareholders for the responsibility of fulfilling our mission and what that's going to require. And I think people should be happy that we are able and willing to not just like race ahead and say, you know, either our business requires this or only we can be trusted with this, or, you know, we just need to be the people that get this right. But to say like, we want to do this with other people, we want to be responsible. We want to be, we want to ensure that no one's taking a 10 % risk, certainly not us, of something really terrible happening.

25:12And that is going to require some unusual things. On the 10 % risk, I just want to be clear because I don't think everybody knows this. P-Doom, like we talked about at the beginning, has been in the industry for a very long time. And correct me if I'm wrong, it seems like to this point, there has not been existential risk. It's what people could fear if this trend continues, the models keep developing of which that could develop. I think that's right. And I think it is when you're living through exponential progress, it kind of like feels flat somehow. And then all of a sudden, everybody wakes up.

25:47It happened during COVID where all of a sudden the world had its like one weekend where things locked down. The progress, and we talked about Navier Stokes a little bit, maybe to put that in context. Three summers ago, we were like pretty good at grade school math. Not even that good. We were like, okay, grade school math. two summers ago we could uh on the aim which is like a pretty good high school math competition you know we could get like a good to a very good score one summer ago we got an imo gold medal which is like the most prestigious math competition barely and this summer um we solved the millennium prize problem this is the summer summer summer if you look at that progress that's like really quite like i just got a little like hair stood up on the back of my neck saying that out loud That's like a crazy thing.

26:34And you project that forward four more times. And that is what I think people are worried about. I do not think that Astra poses any kind of existential risk to the world, obviously. And, you know, I do think our industry has done a little bit of boy who cried wolf, but it comes from a genuine place of care and concern about where this is going. I just want to point out that if you stopped right now, you might have actually achieved your full mission. Like, I feel like you almost have AGI. It's benefiting humanity. There is no risk. Like you are doing pretty well in advertising. You keep cranking on that.

27:07Like, why not just be like, I'm good? I mean, yes, you can always say like, let's be Luddites. You can always say like, we don't want it anymore. We don't want it to be better. Like, I would still like to see the world get much better. I would still like to see diseases get cured and to people that don't have a great quality of life, which many people in the world still don't, to get that. I believe in the potential of this technology to transform people's lives for the better much more than it already has. And I think you can already see some examples of where people are starting new businesses, are doing science, are kind of like transforming their own lives, getting great healthcare advice.

27:46I want much more of that, but we will not put the world in harm's way to get there. We got this. I do take comfort in the fact that you will stop, it sounds like. If you feel like there's a point of no return, you will stop before that happens. I think there will be many places along the way where we have to say we are pausing, we're going to redirect our efforts into safety alignment before we can go to that next level. And this is what we've been doing for a long time. It has worked so far i suspect it will continue to work i don't think we're ever personally going to get to the point we have to say melt all the gpus um but you know if we had to do something like that to ensure the continued existence of humanity easy yes i don't think it's gonna happen i think i think it's just going to continue as like next level of progress next level of safety and alignment requirement like that and there's talk of the collective slowdown need there's only a handful of you guys right it's like you dario elon maybe zuckerberg um sundar or whoever was running google and demis's place why can't you all just get in a room get some beers and figure out how to like solve humanity like just get on the same page i know there's beef but come on i think that will happen like like yesterday i'm not gonna yeah i think part of being like reliable partners.