A proposal to limit how smart machines may become would, in Alexander Wissner-Gross's view, end up limiting how smart people may become. "Because we're going to be merging with the machines, it's effectively a cap on individual human intelligence," said Wissner-Gross, a computer scientist and founder of Reified. He was speaking on a live episode of Moonshots with Peter Diamandis that was recorded on September 25, 2026, and published two days later.
He and his co-hosts were responding to the Ban Artificial Superintelligence Act. Senator Bernie Sanders and Representative Greg Casar announced on September 23 that they had introduced it. The hosts did more than call the bill unwise. They argued that it would limit people as well as machines, that it would stop investment before any rules existed, and that it would not matter once anyone can download an AI model.
What the bill would do
"Superintelligence" usually means AI that outperforms people at most kinds of thinking. According to the sponsors' announcement, the bill would permanently prohibit developing or deploying systems that exceed human cognition across most domains, or that are capable enough to destroy or disempower humanity. It would also pause advanced AI development until a new cabinet-level Department of Artificial Intelligence writes rules and review procedures. That department would monitor dangerous capabilities, oversee their removal and supervise the destruction of prohibited systems. For violations or attempts to get around the law, the sponsors describe dissolving the company and prison terms of up to twenty years. They also propose seeking international agreements and export controls. None of this has passed Congress.
Peter Diamandis, founder of XPRIZE and Singularity University, summed up the plan on air as a permanent ban, an immediate pause and a new department. He described the penalties as a "corporate death penalty" plus 20 years in federal prison, and said they were modeled on the laws against illegal nuclear weapons. The sponsors' own wording is prison terms of up to twenty years. Diamandis called the proposal "fear-mongering at its best."
The hosts linked the bill to a change of vocabulary. At the United Nations General Assembly on September 22, President Donald Trump rejected any global scheme to control AI and said the technology would from then on officially be called superintelligence. Diamandis joked that the United States had "won the race to super intelligence by rebranding it," and that Sanders called for a ban as soon as the new name arrived.
A ceiling on minds
Wissner-Gross made the sharpest case. "Every Hollywood movie needs a villain," he said, and the story of the singularity (the point when machine intelligence races past human intelligence) "has found one now." To him, a legal limit on intelligence is not only "a terrible idea" but "an evil idea," and "tantamount to capping human intelligence." He compared it to the dystopias of science fiction: "burn the books and burn the minds."
His argument depends on a forecast. If people come to think together with machines, a limit on machine capability also limits the people using it. "Penalizing people for being too smart is a terrible idea," he said. He predicted that history would judge any cap on intelligence, "whether it's machine or human or human-machine hybrid," as "basically the equivalent of communism in the 21st century." The sponsors describe a bill aimed at AI systems and the companies that build them. Wissner-Gross's objection rests on his expectation that the line between people and machines will blur.
Fear of a rule nobody has written
Dave Blundin, founder and general partner of Link Ventures, worried less about the ban's goal than about what it would do in the meantime. American liability law, he said, makes it "very easy to scare people into not doing anything." He added: "If you're afraid of 20 years of prison, you don't even know what the rule is. They haven't even proposed a rule." That matches how the bill is built: under the sponsors' plan, the detailed rules would come later from the new department, and the pause lasts until they exist. Blundin expects the result to be "when in doubt, do nothing," which he called "so toxic." He also took a jab at Sanders's home state. He loves Vermont, he said, and "it's beautiful because not much has been done."
Blundin did not dismiss the dangers. "If you do nothing, very, very bad things are going to happen," he said, and named the risks he takes seriously: terrorists using AI to build bioweapons, or China getting "miles ahead" of the United States. His answer was engineering, not prohibition: "very specific technical solutions" for open-source models and for monitoring, "just like with nuclear power." He argued that if work starts now, these problems could be solved "very easily" before superintelligence arrives. Turning the question into a left-right political fight, he said, is "the stupidest thing that could happen right now."
He said most of the damage could be reversed quickly, with one exception: data centers, the warehouse-sized buildings full of computer chips that train and run AI. If they end up in other states, "they're not coming back," he said. "Once the ground is broken, the chips are there, the investment is made, that's the part you'll never get back." He sees these facilities as a "massive driver" of state economies, so he called this the part of the Sanders plan that its supporters would "really regret."
Can a ban reach a downloaded model?
Salim Ismail, founder of Open ExO, began with a question for both camps: what evidence would change anyone's mind? He thinks each side's story is stuck. He then dismissed both the Sanders bill and the Xi–Trump summit as "completely irrelevant" and credited Blundin with the underlying point. A model's "weights" are the numbers that make up a trained AI system. Once those files are published, anyone can copy them and run the model on their own hardware. "The minute you have local weights available and people can run their own models at a local level, the world is going to move ahead," Ismail said. "It doesn't matter what any single person does now."
He welcomed that. "We will likely solve cancer in the next two years," he predicted, and asked, "Why would you want to slow that down?"
The encryption precedent
The discussion also raised a legal objection. Decades ago, the argument went, the US government tried to restrict encryption (the mathematics used to scramble data so only the intended reader can decode it). That restriction was treated as a ban on math and therefore as a free-speech issue. By the same logic, superintelligence is mathematics, and banning it would be unconstitutional.
The history is narrower than that. According to the Electronic Frontier Foundation, which represented him beginning in 1995, Daniel Bernstein wanted to publish his Snuffle encryption algorithm, a paper and its source code, but export rules required government permission. The courts recognized source code as protected expression and rejected the restrictions on publishing it. While the government was still appealing, it revised its regulations and substantially lowered the barriers to publication. Even after Bernstein's win, encryption software stayed subject to export rules, including a requirement to notify the government when publishing its source code. The case concerned the right to publish code. It did not settle whether the government may forbid building and running a system, which is what the Sanders–Casar bill targets.