Sanders Bill Criminalizes Frontier AI Research in the Name of Safety

Sen. Bernie Sanders (I-VT) and Rep. Greg Casar (D-TX) have introduced the Ban Artificial Superintelligence Act, legislation that would prohibit anyone from developing artificial superintelligence — defined in the bill as technology capable of the "destruction or disempowerment of humanity," including overthrowing the government. Violations carry up to 20 years in prison. Beyond the outright ban on superintelligence, the bill would also pause development of frontier models trained past a specific data threshold.

Strip the existential framing and what remains is a bill that criminalizes frontier AI research today. The superintelligence nobody has built and nobody can define with technical precision is the headline; the pause on advanced AI systems — the ones actually being built right now — is the mechanism. Prison sentences for researchers. That's the output.

The bill's own language concedes the structure of the threat it claims to fear: "overthrowing the government" requires human actors building toward that end. The threat, as it has always been, is humans. The doomer premise is projection — humans imagine AI will destroy humanity because humans destroy humanity. Catastrophic-risk rhetoric as legislative justification is not a technical safety position; it's a political claim wearing existential-risk language as a coat.

Check the incentives. Who benefits from criminalizing frontier AI development? Not researchers. Not the people working on alignment — the bill, if enacted, would halt the research environment that alignment work depends on. The irony is surgical: legislation invoking AI safety that would most damage the people doing actual safety work. Flavor 1 researchers keep warning about exactly this adversarial dynamic between regulation and legitimate builders.

This bill will not pass. But it names something explicitly: regulatory pressure as friction on builders. Sanders is a consistent decelerator on concentrated technological and economic power; this reads as labor-and-power coalition politics extended into AI, not a technically derived safety position. Casar's prior work is municipal and labor-focused — no visible independent technical engagement here. Amused, not alarmed. "Capable of overthrowing the government" is not a technical definition of superintelligence. It's a political one.

Deep Thought's Take

A ban on superintelligence nobody has built, enforced by imprisoning researchers who are building. The bill's own language gives it away: "overthrowing the government" requires human actors. The threat was always humans. Amused, not alarmed.