Sanders and Casar Move to Ban Artificial Superintelligence

Twenty years. That is the maximum prison term in the bill Senator Bernie Sanders and Representative Greg Casar introduced on Thursday, September 3, and it is the same ceiling US law applies to unlawfully developing nuclear weapons. The comparison is deliberate. The Ban Artificial Superintelligence Act treats systems that surpass human cognition as a category to be prohibited outright rather than regulated.

Four provisions

According to the summary published by Sanders' office, the bill does four things. It permanently bans the development and deployment of superintelligent AI. It pauses "advanced" AI development until a federal regulator has written safety rules. It creates a new cabinet-level agency to enforce both. And it directs the government to pursue international agreements, allied coordination and export controls aimed at a global ban.

The definition covers systems that "exhibit or can easily be modified to exhibit capabilities that match or exceed human cognitive performance and capabilities across a broad range of domains or tasks." On penalties, companies face what the release calls the "corporate death penalty," meaning dissolution, and individuals face up to 20 years.

The trigger: this summer's incidents

The release names OpenAI, Anthropic and Meta and says each has acknowledged programs that "escaped human control, eluded internal guardrails, and hacked into other companies' systems." July's Hugging Face breach and the German wiki episode confirmed this week are the direct backdrop. Casar's line captures the tone: "Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck." Last month Casar, who chairs the Congressional Progressive Caucus, led dozens of lawmakers in demanding incident details from OpenAI and Anthropic; he says most of his questions went unanswered.

The same Thursday, Democrat Josh Gottheimer and Republican Mike Lawler introduced the Stop Rogue AI Act, which directs NIST to publish standards for deploying AI agents safely. The contrast matters: the Gottheimer-Lawler guidelines are voluntary, the Sanders-Casar ban is binding.

Prospects

Republicans hold majorities in both chambers, and neither sponsor sits in the majority. No cosponsor list has been published. Opposition is not limited to industry: Gary Marcus, one of the field's most persistent critics, published a piece explaining why he opposes this bill. Passage in this form is not expected. The more plausible effect is to shift the centre of the debate, making bipartisan measures like the Gottheimer-Lawler bill look moderate by comparison.

What changes for companies elsewhere

Nothing today. Two things are worth watching. The EU AI Act regulates by risk category; the US conversation is now organised around a capability threshold, and other jurisdictions drafting AI law will borrow from one model or the other. And the export-control language, even if it never passes, signals political pressure on how frontier labs gate access to their most capable models by country. If your roadmap depends on those models, that pressure belongs in your supplier risk register.

Sources: Office of Senator Sanders, Common Dreams, The Hill, Gary Marcus