Sam Altman, the chief executive of OpenAI, said in a Politico interview that "we believe that the world should accept some bad things happening," according to Peter Diamandis, who read the line on an episode of Moonshots with Peter Diamandis. The episode was recorded on October 6, 2026, and published the next day.
Diamandis, who founded XPRIZE and Singularity University, was joined by three regular panelists:
- Dave Blundin, founder and general partner of Link Ventures
- Alexander Wissner-Gross, a computer scientist and founder of Reified
- Salim Ismail, founder of Open ExO
Emad Mostaque, founder of Intelligent Internet, was the guest. The panel's reactions split over why Altman said it and how much risk a society should accept. Blundin said he did not think Altman can guarantee safety.
What Altman said, as read on the show
Diamandis summarized Altman as arguing that technology should remain broadly accessible and that people should have agency. He said Altman described a basic difference in worldview between OpenAI and its rival Anthropic. Altman wanted a "lighter touch" on regulation, while Anthropic wanted strong regulation.
Diamandis then read a longer passage. Altman said he understood the view of people who think the technology will become so powerful and dangerous "that a single lab in San Francisco should have it and make sure nothing bad happens" while doling out the benefits. He called that "a completely unacceptable tradeoff." The passage continued with a list: no major hacks, no misuse, "zero scams."
Diamandis called that statement "pretty extraordinary" and asked how Altman could possibly promise it. But in the order Diamandis read it, the zero-harm list seems to describe what a single controlling lab would promise, which is the arrangement Altman had just rejected. That makes it less a guarantee from Altman himself than a description of the bargain he refuses. As Diamandis read it, Altman ended by saying people would do "orders of magnitude more good stuff than bad stuff."
Diamandis also recalled that Eric Schmidt, who has appeared on the show, had said he expected and hoped for a small catastrophe that would wake people up before a big one.
The PR reading
Blundin saw the remarks mainly as positioning. He described a "PR battleground" among AI's biggest figures. He named Elon Musk, Dario Amodei, Altman, Donald Trump and Jensen Huang. People he surveys, he said, like one in five of them "at most". Demis was his exception: "everybody loves Demis, actually."
In Blundin's account, Amodei, Anthropic's chief executive, had taken the moral high ground by pointing to more pausing, more safety work and more testing, and then saying models would be kept internal because they were too dangerous to release. That opened the door, Blundin said, for Altman to win the high ground back with a message about access for everyone. "I don't think Sam can guarantee safety. And of course not," Blundin said. But Altman could promise to give listeners "the best of the best tools" so they are not deprived. "Dario is going to try and deprive you," Blundin said, calling it "a really smart message."
The end of the "cartel"
Wissner-Gross connected the remarks to a debate from a few weeks earlier. In September, Amodei published an essay, We Must Pace the Frontier. It proposes giving outside evaluators deep, ongoing access to a lab's models and operations. Frontier labs in democratic countries would coordinate safeguards with government help, with attention to antitrust law, which normally limits coordination among competitors. Governments would also seek verifiable international coordination, including with China. Amodei presents pacing as time for alignment, evaluation and safeguards, not as abandoning development.
Wissner-Gross said the essay was "arguably" a call to form an AI safety cartel. "Cartel" is his word, not Amodei's. He recalled that Altman's response at the time was "I'm in. Sign me up." Altman's new comments, Wissner-Gross said, were "hopefully a reflection that that plan is dead. The cartel is over." He allowed that, read generously, it might be "some triangulation on Sam's part."
Wissner-Gross said bad things happening from time to time are "the cost of freedom and the cost of competition." Diamandis added: "The cost of progress." Wissner-Gross compared AI to cities: some crime will occur, but that is "definitely not" a reason not to form cities. "Every concentration of intelligence will come with a benefit and a downside," he said. He welcomed Altman's "far more realpolitik note and tone."
Promise and peril
Ismail agreed with Blundin's PR framing. He also said every technology in history has two sides, and cited Neil Jacobstein's question of how to "extract the promise without the peril." Cars kill people and electricity electrocutes people, he said, yet neither is banned. The job is to build guardrails. He argued that developing countries are a reason to get the technology out "as fast as possible," because it could lift people several levels of Maslow's hierarchy of needs. "We've done it very successfully throughout history," he said. "Why do we think this time we won't do it?"
Diamandis argued that regulation is written after problems appear. Federal aviation regulations, he said, are "written in blood" because a rule follows each accident. AI would go the same way: push the boundaries, find the problems, then regulate. Ismail said the goal should not be zero bad outcomes, since "that means you just ban the technology." It should be orders of magnitude more positive outcomes than negative ones.
Risk is not a single knob
Blundin said anyone with an engineering mindset should reject "the idea that there's a big knob labeled risk and we need to dial it left or right." Good engineering, he said, can cut risk to "near zero" levels while keeping the full benefit. Society did not do that with nuclear power; it treated risk as a knob and dialed progress to near zero, which he called "the stupidest thing." He said the same thing was about to happen with AI, an "engineering problem with engineering solutions."
Ismail pushed back. Open-weight models, whose files anyone can download and run, are already out there, he said. Governments can regulate outcomes, guardrails and the "on-ramps and off-ramps," but "you can't regulate the core technology."
Wissner-Gross said it could be done. He imagined a nightmare like American Prohibition, with "bathtub gin" supercomputers and people using odd hardware for AI training or inference. Authorities could confiscate all the GPUs and TPUs (the specialized chips AI runs on), or, in his words, "as Eliezer is constantly calling," bomb the data centers. "But I think it would be a terrible, terrible world to live in," he said.
Mapping what could go wrong
The conversation then turned to how people decide under uncertainty. One argument was that most frightening scenarios, such as building viruses or launching cyberattacks, are already regulated. The newer difficulty is the move from deterministic software to more flexible agents. When people weigh outcomes by probability, that is an expected-utility calculation. Under uncertainty, the argument went, people instead use "regret minimax": they minimize their maximum regret. Fear stories push toward that approach, which yields choices like no nuclear power stations and no superintelligence. The proposed remedy was a mapping project that sets out, in their own words, what the scared and the positive each say about every scenario. That would show which risks need a policy or regulatory response and which need to be communicated to the public.
Diamandis closed with an aviation example. In the Reagan years, he recalled, Transportation Secretary Elizabeth Dole announced a goal of zero aircraft deaths, and "there was only one way to make that happen, stop flying airplanes." Wissner-Gross applied the same logic to AI: if the world wants zero danger from superintelligence, the solution is to "just ban intelligence altogether, human and otherwise."