Bernie Sanders Proposes Banning Artificial Superintelligence

Bernie Sanders and Texas Rep. Greg Casar announced a bill today that would make it illegal to build artificial superintelligence in the US. It would freeze a lot of advanced AI development until the federal government can create a new agency that would be responsible for regulating AI and creating safety rules.

The bill is called the Ban Artificial Superintelligence Act. No person or company would be allowed to develop or deploy a “superintelligent” system. Advanced AI would be paused until the new regulator exists and has a model review process. Companies that try to get around the law could be shut down entirely. Individuals could face up to 20 years in prison. Sanders is comparing that to penalties for illegally building nuclear weapons.

Their definition of superintelligence is an AI that matches or exceeds human intelligence across a wide range of tasks, or one capable of planning to take humans out of power. They are concerned about AI undermining the US government, or ignoring shutdown commands.

The agency would monitor frontier systems for their entire lifespan, order dangerous capabilities removed, and supervise destroying a model if it crosses that line. An advisory board of AI experts would sit with it.

This is the hardest version of an argument Sanders has been making all year. In August he told the CEOs of OpenAI, Anthropic, and Meta to pause, citing their own warnings that they might lose control. He already filed a moratorium on new AI data centers. He also tried to move 50% of the big labs’ stock into a public wealth fund. Today is the version that just outlaws the end state machine.

I understand the instinct. The same executives give interviews about existential risk and then ship the next model. If you take that risk seriously, hoping the market polices itself is not a strategy.

The bill still has a hole in the middle of it.

“Superintelligence” is the whole fight, and the definition is vague on purpose. Current models already write, reason, and pass professional exams. Next year’s models will do it better. A statute that bans anything matching “human cognitive performance across a broad range of domains” will not land as a clean cutoff. It will land as years of lawyers arguing about the semantics of AI while labs take their business elsewhere.

The pause is the other problem. It lasts until a new federal department exists and has rules. Without that infrastructure in place when the bill passes, the US will lag behind in the global technology race. We all know how quickly the government moves…

I do believe that AI could cause some issues without any guardrails. The line between completely hindering progress and keeping control of the product is thin. When the founders and CEOs of a product express safety and control concerns, we should listen. This bill is too extreme, though. Congress should have a frank discussion with the current frontrunners in AI technology about where they see it going and what they believe could be a threat in the future. We don’t need to jump too fast and stunt our technological advancement potential. Stalling progress could be the difference between military power, counterintelligence, business competitiveness, etc. with our adversaries.