Sanders wants to ban superintelligence by law: the problem is that nobody can define it

By Zendoric · September 24, 2026.

Independent Senator Bernie Sanders (Vermont) and Democratic Representative Greg Casar (Texas) introduced a bill in the United States Congress on Wednesday that would ban the development of "artificial superintelligence"—the hypothetical level at which a system would surpass humans at virtually any cognitive task—and create a new federal agency, dubbed the Department of AI, to oversee the most advanced systems. That is how WCAX and The Seattle Times describe it in nearly identical terms, framing the announcement at a moment when, according to the report itself, some of the AI industry's top executives are themselves asking Congress to impose guardrails on the technology.

That the bill does not merely "regulate" but "ban" is in itself a statement of intent. Sanders has spent years warning about the effect of automation on employment and on the concentration of corporate power, a concern he has translated into earlier proposals on sharing work and its benefits. Casar is one of the most visible voices of the progressive wing of the House of Representatives. That the two jointly sign a text calling for an outright ban, rather than one setting gradual limits, places the initiative at the hardest end of the regulatory spectrum now being debated in Washington.

Another report on the same announcement, from Gizmodo, links the measure to the rhetorical shift of the Trump Administration, which has reportedly begun using the term "superintelligence" in its own discourse on AI. If that reading is correct, the Sanders-Casar bill would be both a legislative proposal and a political rebuttal to the language the White House has chosen to describe its technology plans, rather than a merely technical gesture.

The bill's ambition should be separated from its real chances. With a Republican-controlled Congress and a White House focused on expanding the country's AI infrastructure to counter China, a law that bans an entire category of technology outright has, today, close to zero legislative prospects. It is, above all, a positioning tool: it stakes out a recognizable stance long before it has any chance of becoming law.

But the problem is not only parliamentary arithmetic, it is one of definition. Neither the scientific community nor the AI labs themselves have agreed on what capability threshold separates an "advanced" model from a "superintelligent" one: there is no metric, compute level or common benchmark that marks that boundary. Legislating a ban on something that cannot be precisely measured is risky: either it remains a symbolic gesture that is hard to enforce, or it is drafted so broadly that it can sweep in beneficial research—in drug discovery, for example—that no one intended to halt.

At Zendoric we have argued that the real governance problem is not whether AI advances, but who decides its rules and on what evidence. From that vantage point, the most usable part of the Sanders-Casar bill is not the ban but the other leg: a dedicated federal agency with the technical capacity to continuously monitor frontier models, closer to what the FDA already does with drugs or the FAA with aviation than to a blanket ban on an undefined concept. That model—an expert agency, continuous review, measurable criteria—is defensible. Banning by decree a technology that not even its own creators can delimit, far less so.

That it is industry executives, and not only activists or unions, demanding guardrails confirms something we have already noted: business and risk have stopped being perceived as opposing forces within the sector. That does not make those requests disinterested—asking for regulation also serves to set the playing field in favor of whoever already dominates the market—but it does confirm that concern about the pace of deployment has moved beyond the outsiders.

Over the long term we continue to believe AI's trajectory points toward a society with less disease, better health and greater material abundance, and none of that changes because a senator proposes banning a still poorly defined word. But the short term remains hard, and the job losses and concentration of power that worry Sanders are real, not a rhetorical bogeyman. The most useful response to that fear, however, is not to ban what we still cannot name precisely: it is to build the institutional capacity—measurable, technical, permanent—to monitor it closely as it arrives.