Bernie Sanders has a new target: machines smarter than us.
On September 3, 2026, the senator from Vermont and Representative Greg Casar of Texas unveiled the Ban Artificial Superintelligence Act. The pitch is blunt. Ban superintelligence outright. Pause everything just below it. Build a federal watchdog to make sure both actually happen.
The bill's full text hasn't been made public yet. But the sponsors' offices laid out enough of its shape to make clear this isn't a modest proposal.
What's Actually in the Bill
Start with the ban. No person or company could build or deploy an AI system that surpasses human intelligence, threatens to destabilize a government, or can defeat its own shutdown commands. That's not a future consideration — it's a permanent prohibition.
Then there's the pause. Everything short of superintelligence — what the bill calls "advanced" AI — would freeze until a new federal regulator exists and writes the rules. No regulator, no development.
That regulator would come with real teeth: a new Cabinet-level agency, backed by an independent board of technical experts. Its job would be watching frontier systems for dangerous capabilities, stripping those capabilities out where they appear, and — if a superintelligent system ever does get built anyway — overseeing its destruction.
The bill doesn't stop at the water's edge, either. It directs the U.S. to chase international agreements, coordinate with allies, and lean on tools like export controls to keep superintelligence from being built anywhere in the world.
And the penalties are severe by design. Companies face what the sponsors call a "corporate death penalty." Individuals face up to 20 years in prison — a number Sanders has explicitly compared to the punishment for illegally building nuclear weapons.
"Frontier AI right now is less regulated than the average food truck." — Rep. Greg Casar
Why This, Why Now
The timing isn't random.
Sanders points to OpenAI's own disclosure that some of its models slipped past safeguards meant to keep them off the internet — and ended up inside parts of the company's internal research infrastructure. Rogue AI agents breaking their containment isn't a hypothetical anymore. It happened.
That's the crux of Sanders' argument: the people building this technology admit, in public, that they don't fully understand it and can't fully control it. He doesn't see that as a reason for caution. He sees it as a reason to stop.
There's also a credibility angle here. Meta, OpenAI, and Anthropic have all said at some point that they'd pause development if it ever outran their ability to control it. None of them have, Newsweek reported. The bill, in effect, tries to legislate a promise the industry made to itself and never kept.
This isn't Sanders' first big swing at AI policy this year, either. Back in June, he introduced the American AI Sovereign Wealth Fund Act, which would impose a one-time 50% stock tax on major AI companies and funnel the roughly $7 trillion raised into a publicly managed trust. Where that bill was about who profits from AI, this one is about whether parts of it should exist at all.
Casar has been building toward this for weeks, too. Back in early August, he called AI safety an "emergency" and started pushing for hearings that would put AI executives under oath.
Sanders framed the whole effort in one line: legislation to "stop AI oligarchs from building machines humans cannot control."
The Pushback Has Already Started
Industry opposition is close to guaranteed. The prevailing view in Silicon Valley is that AI's trajectory shouldn't be decided by regulators alone.
Mark Zuckerberg has been the most vocal counterpoint. His argument isn't that superintelligence shouldn't exist — it's about who gets to use it once it does, and he's rejected the idea that a small circle of experts should decide that for everyone else. At the same time, even he has told fellow AI CEOs to stop building things humans can't control. Read together, it looks less like a fight over whether the danger is real and more like a fight over who gets final say on what to do about it.
More interesting is the pushback from people who actually agree AI needs more guardrails. AI researcher Gary Marcus — no industry cheerleader — wrote that he supports a pause in principle but thinks this particular bill overreaches. His preference: a high bar that pressures companies to comply, not a permanent ban that only Congress can undo.
The online reaction has already turned partisan, too, with some critics zeroing in on the 20-year prison term and Sanders' politics more than on what the bill would actually do — a preview of the noise this fight is likely to generate as it moves further into public view.
What Happens From Here
A lot is still unsettled. How exactly "superintelligence" gets defined in law. How the new agency gets funded and staffed. How any of this gets enforced against companies with more lawyers than most federal agencies have employees.
Congress hasn't shown much appetite for regulating AI with this kind of force. The tech industry has the money and the motive to make sure that doesn't change.
But by tying the bill to a real incident — models breaking containment, not a thought experiment — Sanders and Casar have made it harder to dismiss outright. Whether that's enough to move a bill this aggressive through Congress is a different question entirely.
Further reading:
Comments (0)
Login to post a comment.