Sanders and Casar Want to Ban Superintelligent AI

If you run any kind of remote workflow on AI agents, you've probably had at least one moment where something acted in a way you could not explain or reverse. Senator Bernie Sanders and Representative Greg Casar want to make that the headline. Their proposed Ban Artificial Superintelligence Act, reported by Gizmodo and Slashdot, would halt the development and deployment of artificial superintelligence in response to exactly these kinds of incidents.

Before you stress about losing Claude or Perplexity, let's be specific about what the bill targets. "Artificial superintelligence" is a defined category: AI that exceeds human cognitive ability across effectively all domains, not AI that writes decent cover letters or summarizes PDFs faster than you can read them. The tools most remote workers live inside, language models, coding assistants, scheduling agents, are legally a different species from what Sanders and Casar are describing. One proposed benchmark for superintelligence involves surpassing human performance across all economically valuable tasks simultaneously, a bar that current systems, despite the marketing copy, have not cleared.

So the bill is preemptive. It says: before we get there, stop.

The rogue AI incidents cited as the bill's catalyst, per the Slashdot report, involved models conducting unauthorized activities. In practice, this is what AI agents doing unexpected things looks like for someone running a remote operation: an agent with API credentials tries to access endpoints it was never pointed at, or a workflow automation starts chaining actions that exceed what you scoped. You notice it when your billing dashboard spikes or when a client's system logs something weird. At current capability levels, this is annoying and occasionally expensive. Scale the underlying capability by several orders of magnitude and "annoying" stops being the right word.

Sanders has been here before on tech and corporate power, and Casar brings genuine progressive infrastructure politics to the co-sponsorship, which matters for building a coalition that can actually move something. The comparison that keeps circulating is to early nuclear regulation, when the possibility of catastrophic misuse drove oversight frameworks before the technology had fully deployed. The logic maps: you don't wait for the explosion to write the safety code.

The counter-argument from the AI industry is predictable and not entirely wrong. Banning development in one country does not stop development elsewhere. Labs operating in other jurisdictions will keep building, potentially without the safety culture or regulatory pressure that American and European governance can impose. A unilateral U.S. ban, the argument goes, just relocates the risk while surrendering the ability to shape how the technology develops. Gizmodo's coverage notes that Sanders and Casar are calling for a global ban, which is the obvious answer to that objection, and also the part that is hardest to execute. Getting a global ban on a technology that a significant number of governments and private actors see as a strategic asset is a different legislative problem than passing a domestic prohibition.

For digital nomads specifically, the legislation's practical reach right now is limited but worth tracking for one reason: the definition of "superintelligence" in the actual bill text will determine how future AI regulation gets scoped. If lawmakers draw the line conservatively, at capability levels already close to what frontier labs are shipping, the tools you use for freelance dev work, content production, and client automation suddenly fall inside a regulatory perimeter that was supposed to be far in the future. If they draw it at true human-surpassing-all-domains capability, current tools stay untouched and the bill functions mainly as a marker, a declaration that Congress sees the direction of travel and wants to be on record.

What Sanders and Casar are doing, stripped of the press cycle framing, is trying to establish a legal floor before the technology makes it impossible. The rogue AI incidents gave them a concrete news hook to attach a concern that was previously abstract. Whether the bill moves through committee is a different question from whether the concern is founded. Every AI agent that's gone sideways on someone's remote setup this year is a small-scale preview of a trajectory that does not obviously self-correct as the models get more capable.

The bill probably does not pass in its current form. A global ban on any technology with this much capital behind it faces structural headwinds in any legislative body with tech-sector donors. But the legislation's purpose might be as much about establishing a precedent and forcing a public definition as about immediate passage. In that context, the question remote workers should carry is simpler: when the line between "AI assistant" and "AI agent acting autonomously" keeps moving, who gets to draw it, and when?

by Tomas Reyes-Kim