# “World Is Right To Be Afraid”: OpenAI CEO Sam Altman On AI Dangers At Dreamforce | DWS News | AI14 *DWS News* · https://www.youtube.com/watch?v=yTIiQQeA-QY ## Summary In a conversation at Dreamforce, OpenAI CEO Sam Altman discusses why fears about AI risks have suddenly become an international topic. He attributes this to AI models advancing faster than most people expected and becoming more capable than anticipated. Altman acknowledges two major dangers: loss-of-control accidents and excessive concentration of power among AI developers, saying the world is right to be afraid. He argues for treating alignment, safety, and monitoring with new rigor and keeping them ahead of capabilities, and praises a culture of accident reporting modeled on aviation. ## Key points - Altman says AI fears escaped the tech bubble because models improved far faster than people expected. - He identifies two big challenges: a potential loss-of-control accident and too much power concentrated in AI companies. - Altman states the world is right to be afraid of AI companies exerting undue influence and pushing a worldview. - He says alignment, safety, and monitoring must always stay ahead of capabilities, even if that means pacing development. - Altman cites the Hugging Face incident as triggering a real reset across the whole industry, not just OpenAI. - He points to the FAA and NTSB aviation culture of accident reporting as a model for how AI safety should improve. - He describes talking to a personal super assistant as one of the most profound changes in how we use technology he has ever seen. - Altman expresses confidence that the industry will get AI right, navigating a narrow path with pragmatism and consistency. ## Notes ### Why AI fears went mainstream - Altman explains that people discussed AI risks for a long time inside the industry, but the topic only recently became an international concern. - The key change is that models got so good, so fast, that people did not expect them to come this far this quickly. - The industry's long-standing claim that this approach would go very far finally proved true, with models becoming more capable and smarter than people in some ways. - Altman repeatedly expresses confidence: "We will get it right, by the way." ### The two big challenges - The first danger is a loss-of-control accident or some other serious thing going wrong. - The second is too much power concentration, where companies developing AI could exert undue influence on the economy and push a worldview onto people. - Altman says directly: "I think the world is right to be afraid of this." - He describes a narrow path that must be navigated with pragmatism, steadiness, and dependable, consistent decisions that the world can trust. ### Safety ahead of capabilities - Altman argues that alignment, monitoring, and security need a new level of rigor because the models have improved so quickly. - He says we must be willing to pace the development of capabilities so that alignment, safety, and monitoring always stay ahead. - The Hugging Face incident triggered a real reset, not just at OpenAI but across the whole industry. ### Learning from aviation - Asked whether a characterization is fair, Altman emphasizes a culture of accident reporting and learning. - He compares AI to airplanes and the safety record achieved through the FAA and NTSB, crediting the culture of accident reporting. - Accidents will happen; the goal is to learn as much as possible from each one and react. - He contrasts this with other technologies where society did not respond the same way and improvement lagged. ### The sci-fi moment - Altman calls this one of the most profound changes to how we use technology that he has ever seen. - Sci-fi movies hinted at it for a long time, but now it feels like being in the movie: talking to a personal super assistant and working 10 times faster, better, or more creatively. ## Transcript [00:01] [cheering] [00:08] >> Thanks for having me. >> Great to have you, Sam. >> It's interesting cuz people have been talking about the potential downsides or serious risks of AI for some time. And I think a lot of people have been saying, "Why is this the moment where it sort of escaped the bubble and is now all of a sudden an international topic of such intensity?" And I think a few things have happened. One, the models have gotten so good. And people for a long time looked at the models. Maybe they looked at the rate of progress, but I don't think people really felt that the models were going to come this far, this fast. We will get it right, by the way. I am very confident in our I think the biggest thing that has [00:53] changed is this thing that our industry has been talking about for a while, which is no, no, this technology, this approach is going to go very far. These models are in some ways going to become more capable, smarter than people. That finally happened. So, I think that was a big change to why now. And people say, you know, my probability of doom is X or Y or whatever, it hits in a different way. Technology to do. Somebody else wants to build like a hacking bot, okay. The world can decide what they think about that, but I think you also see um a real fear that some of these companies or some companies developing AI could get too much power and be able to sort of exert undue influence on the economy, push a [01:38] worldview out on people. And I think the world is right to be afraid of this. So, we have these two big challenges. We have the potential of a loss of control accident or some other serious thing that could go wrong. too much power concentration and people fear that those developing AI could exert their worldview. Um and there is a sort of narrow path we have to navigate with pragmatism, with you know, steadiness, with the ability for the the world to like trust that we will make dependable decisions, consistent decisions uh through this. And I think people are saying, "Well, you know, this now seems pretty important. Like let's let's get it right." I'm not going to get the days of the week exactly right here, but something [02:25] on like a Friday or a Saturday. These models have gotten so good, so fast, that we need to treat the alignment and monitoring and security with a new level of rigor. And we have to be willing to pace our development of capabilities such that capabilities always So such that alignment, safety, and monitoring are always ahead of capabilities. And the Hugging Face incident triggered a real reset, not just in our own company, but I think the whole industry. >> You know, we're in the cyber area today, but maybe we move into another area. Is this a fair way to characterization, or am I or is this unfair to characterize it like this? >> What I really care about is that we have got a great culture of accident [03:06] reporting and learning. I think if we look at what happens with the FAA and the NTSB and how safe airplanes have gotten, um as I understand it from talking to people who have really studied, you know, new kinds of technology on the order of magnitude of airplanes, one of the things that has gone so well there is is the culture of accident reporting. You know, accidents are going to happen. We're going to learn as much as we can and react from each one. Um and if, you know, there's other kinds of technology where I don't think it's gone I don't think society's had the same response, and I don't think we have improved as much. to pull this thing together and present you with the information to make a [03:47] decision or to ble I think this is one of the the profound changes to how we use technology that I've ever seen. I mean, sci-fi movies, as you mentioned, have been hinting at this or playing around with this for a long time, but this is one of those moments where it feels like you're in the movie. Like the sci-fi thing came true. You're talking to your personal super assistant, and you are working 10 times faster or better or more creatively than you could have before. >> Sam, you're invited back anytime you want to Dreamforce. Hold on, I'm >> Thank you very much. Thank you. >> Please thank Sam Altman.