Sam Altman Rarely Opens Up: 10% to 20% Chance AI Destroys Humanity, Yet OpenAI Still Moving at Full Speed
In a recent detailed interview, OpenAI CEO Sam Altman gave a number for the risk that artificial intelligence (AI) could destroy humanity. He said there is about a 10% to 20% chance. This was the first time he had publicly given such a specific estimate. But Altman is also pushing OpenAI to build more powerful AI models as quickly as possible and keep adding more computing power. This has brought attention back to a difficult question: how fast should AI improve, and what safety limits should be in place?
Altman shared these thoughts during a 45-minute interview with AI blogger Matthew Berman. He also said today's AI models are not conscious in the way people usually understand consciousness. Even if AI develops some kind of consciousness in the future, he believes it would be very different from human consciousness. He emphasized that AI should help people, not decide humanity's future for them. As he put it, "This technology is not a person, and it is absolutely not superior to people."
Altman is not the only major technology leader who has given this estimate. Elon Musk has said several times in public that AI has a 10% to 20% chance of ending human civilization. It is unusual for two such influential technology leaders to agree on the same number. However, they are taking different paths. Altman leads a company racing to develop artificial general intelligence (AGI), which means AI that can perform a wide range of thinking tasks at a human level or beyond. Musk has started his own separate AI company.
Why Altman Believes AI Should Start Early but Improve Slowly
Altman strongly believes AGI will become real, and developing it is part of OpenAI's official mission. In a 2023 blog post called "Planning for AGI and beyond," he explained an approach he called "short timelines and slow takeoff." This means starting to develop AGI soon but allowing its abilities to grow gradually. His reasoning is that if AGI arrives sooner, there will be fewer computing resources, such as GPUs, available to train it. GPUs are powerful computer chips used to build AI systems. With fewer resources, it may be harder for AGI to suddenly become artificial superintelligence (ASI), an AI far smarter than humans. But if development takes longer, more unused computing power could build up. That might allow AI abilities to grow very suddenly later.
Altman explained this idea further in an article published in September 2024. At that time, he said superintelligence could arrive within a few thousand days. He also predicted that AI could make solving climate change, settling in space, and discovering all the laws of physics "become routine very quickly." In early 2025, he went even further, saying, "we already know how to build AGI." But back in 2015, he had described superintelligence as "the greatest threat to human existence." His concerns about danger have not gone away. Instead, he believes faster development can exist alongside efforts to manage those dangers.
During the interview, Altman also discussed a more controversial idea called Recursive Self-Improvement (RSI). This is when an AI system keeps improving itself, making it more capable each time. Those improvements could happen faster and faster. Altman said that if someone pushed RSI forward as quickly as possible, this kind of rapid improvement could really happen. But he warned that "continuing to do that is very dangerous." He believes the world needs some kind of "speed limit" to control how quickly RSI develops. This was one of his clearest public warnings about the idea.
OpenAI Wants Far More Computing Power, and Its Own Chips Are Especially Important
When discussing OpenAI's business and computer systems, Altman was just as ambitious as before. Someone asked him how many GPUs the company would need in the future. He answered, "Infinitely more. However many I can get, that's how many I want."
Altman said that people discussing the worldwide GPU market often miss something important. Most estimates assume AI will stay about as capable as it is today and that the cost of completing each task will stay the same. But if powerful AI becomes much cheaper to use, many more people and businesses could use it. That would make it difficult to put a limit on how large the market could become. Demand could keep growing even as AI becomes more efficient. This is similar to an economic idea called the Jevons Paradox, where making something cheaper or more efficient can lead people to use more of it. Altman admitted there will probably be times when there are too few chips and other times when there are too many. Still, he believes the overall demand for computing power will keep increasing over the long term.
OpenAI is working on several ways to prepare for this growing demand. Altman said developing the company's own computer chips, including future generations of those chips, is the single most important part of its plans. OpenAI wants these chips to use much less energy for the amount of work they perform than today's products. The company also plans to use robots to make equipment racks and build data centers more efficiently. It wants to invest in different parts of the supply chain as well. Altman imagines a future where humanity can build computing systems much more cheaply and get much more computing performance from each watt of electricity. But reaching that goal would require several different technologies to improve successfully at the same time.
Altman also talked about competition. He said OpenAI faces pressure from competitors "almost everywhere," but he believes that is good for users. He was asked what would keep OpenAI ahead if competing AI models eventually became equally capable. His answer focused on familiar business advantages. Strong relationships with customers would still matter. So would products that people and businesses use every day, the benefits that come when more people use the same service, and the data a company has collected over time. Having the biggest AI computing systems that can operate at the lowest cost would also give OpenAI a major advantage.
AI Has Not Replaced Programmers — They Are Busier Than Before
One of Altman's most surprising observations was about computer programmers. Many people expect AI tools that write code to replace software developers. But Altman said he is seeing something different. Programmers are busier than ever, have more influence, and may even be making more money. He believes this is happening because the world needs much more software than people previously imagined.
Altman said something similar is happening in mathematics. Some mathematicians do not like the idea of AI solving mathematical proofs by itself. But Altman argued that a very complicated proof is not always useful if people cannot understand it. New work is needed to understand the ideas AI discovers and explain them in ways ordinary people can read. He admitted that the meaning of "being a mathematician" could change greatly between last year and next year.
Altman said he has experienced this strange situation himself: technology makes work more efficient, but people become even busier. He revealed that he has not looked at a line of computer code for about a year and might never write code himself again. He suggested that writing code when he is older might simply be something he does to remember the past. He understands that solving difficult problems and writing beautiful, well-designed code can make people feel happy and proud. If AI takes over those tasks, people may lose some of the enjoyment that comes from creating and discovering things themselves. However, he argued that "we shouldn't delay the discovery of cures for disease just to preserve that joy of research."
Altman expects AI to discover its first truly important possible treatment for a major disease in about two years. However, he said testing the treatment and getting approval from the US Food and Drug Administration (FDA) would still take additional time. Although he is excited about AI helping medicine, the discovery he personally looks forward to most is a major breakthrough in physics.
OpenAI Needs New Devices, but People Must Stay in Control
Altman also explained why OpenAI needs to create new devices for everyday users. He said this is not because he is worried about depending on other technology companies. He described Apple as a responsible company that manages its platform well and said he is not concerned that Apple will remove ChatGPT. Instead, the problem is that today's computers and phones were designed for humans, not AI. Computers use a huge amount of processing power to create pictures, buttons, and menus for people to see. Then AI has to spend effort understanding those same screens. Altman believes this is an awkward and inefficient way for AI to work with technology.
He imagines a personal AI assistant that can continuously understand what is happening around it and take action before someone has to ask. It would understand the situation even when the person is away from a computer. Altman mentioned an internal tool he uses that saves him about one hour every day. He said it has also reminded him several times about important tasks that needed urgent attention. He believes most of the technology needed for this kind of assistant already exists. What is still needed is work to turn it into a finished product, more powerful computing systems, and new hardware that OpenAI plans to release at some point.
When discussing government rules and ethics, Altman said society needs to agree on one important principle: the future must remain centered on humans. He is concerned about arguments that begin by asking whether AI is conscious and then suggest that "humans perhaps should not always be in control of the future." In his view, treating AI models as if they are people is a dangerous place to start. He worries that believing AI should be treated like a person could eventually lead people to make that belief a reality.
Altman also shared his view of how people use AI today. He estimates that only about 1% of people use AI in a truly advanced and skilled way. Most people have not yet learned to take full advantage of what it can do. He hopes AI will do more than create another industrial revolution that makes people work harder and stay busier. He wants it to bring a renaissance, a period of new ideas and creativity, that gives people more freedom and more opportunities to create things.
Altman's comments show a clear conflict. On one hand, he believes there is a 10% to 20% chance AI could destroy humanity. On the other hand, he wants OpenAI to keep expanding as quickly as possible, saying he wants as many GPUs as he can get. He recognizes the danger but is not slowing down the company's overall development. For investors, this suggests that computing power, computer chips, data centers, and energy will remain some of the most dependable areas for investment as AI grows. For governments and regulators, questions about setting a "speed limit" on AI systems that improve themselves and deciding whether AI should ever be treated like a person are becoming practical problems that may need real rules, rather than just ideas for philosophical debate.