Bill Seeks Permanent Superintelligence Ban in the US

Sen. Bernie Sanders and Rep. Greg Casar formally introduced the Ban Artificial Superintelligence Act in the Senate on September 23, turning an announcement from earlier this month into actual legislation. The superintelligence ban would permanently prohibit the development and deployment of superintelligent AI and temporarily pause advanced AI development until a new federal regulator establishes safety rules.

It would create a cabinet-level Department of Artificial Intelligence to monitor frontier AI laboratories, strip out dangerous capabilities and enforce the superintelligence ban. The legislation also directs the United States to pursue international agreements so that superintelligent systems are not built anywhere else, according to the sponsors' announcement. That international push runs parallel to a Senate letter urging Trump to negotiate AI guardrails with China's Xi Jinping during their September 24 summit. The full press release from Sanders' office describes the measure as an effort to stop what it calls AI oligarchs from building machines humans cannot control.

What the bill would do

The superintelligence ban targets a defined threshold: AI systems that exceed human abilities on most tasks, or that would be capable of destroying humanity. It does not cover current chatbots, coding assistants or consumer agents, according to an analysis of the bill text by explainx.ai. The broader and vaguer piece of the proposal is the pause, which would stop advanced AI development until a federal agency with a licensing regime is fully operational.

Penalties under the superintelligence ban are set on the scale of existing nuclear weapons violations. Individuals could face up to twenty years in prison, while companies would face what the sponsors call a corporate death penalty, the dissolution of an offending entity's charter. The new department would supervise the removal of dangerous capabilities from frontier systems, such as subverting shutdown commands or conducting unauthorized cyberattacks, and in extreme cases the destruction of artificial superintelligences, according to the sponsors' summary.

Sanders framed the superintelligence ban as an answer to a run of incidents in which the largest AI labs appeared to lose control of their own technology. In a statement, he pointed to "frightening new" stories appearing nearly every day about systems "escaping their control" and said it is irresponsible for society to let the companies keep building more advanced products. The future of humanity "cannot be left in the hands of a handful of Big Tech oligarchs," he said, arguing that development should pause and that systems humans "cannot fully control" should be banned both at home and around the world.

Why the timing mattered

The announcement landed on September 3, the same day OpenAI president Greg Brockman used the GPT-6 Astra launch briefing to tell reporters, "Welcome to the AGI era." One of the year's biggest model launches and a proposed superintelligence ban arrived within hours of each other. That same day, the ARC Prize organization, which maintains the ARC-AGI-3 benchmark, independently scored Astra at 62.7 percent, a wide gap from the 98.6 percent figure OpenAI reported using its own evaluation harness, as reported by Tech Times.

The collision was not accidental, Tech Times reported. While Brockman was presenting Astra as the arrival of artificial general intelligence, Sanders and Casar chose that exact day to propose making the development of such a system a federal crime. Casar sits at the center of a parallel channel as well: he received a letter from OpenAI committing to build "automated shutdown" capabilities for its AI agents, according to an analysis by FourWeekMBA, a voluntary step that stands in contrast to the superintelligence ban's criminal penalties. US-China AI diplomacy has also included a proposed AI notification hotline between Washington and Beijing, floated during Treasury talks in New York. Background on the bill's history and provisions shows the measure was announced in early September and formally introduced in the Senate later that month.

What comes next

The superintelligence ban has picked up backers from unusual corners. The ACLU and the Electronic Frontier Foundation support it, and two industry insiders, OpenAI researcher Juan Felipe Cerón Uribe and Google DeepMind engineer Swante Scholz, have endorsed the measure, according to a report by The Jo AI.

Its legislative prospects are thin. The bill faces a Republican-controlled Congress and has no committee movement yet. Betting markets give any American AI safety bill, including this superintelligence ban, roughly a 13 percent chance of passing before 2027, according to explainx.ai. Most observers treat the measure as a message bill, one meant to force a question into the open rather than to pass in its current form.

That question is getting harder to avoid. A second, narrower bill introduced the same week, the Stop Rogue AI Act from Reps. Josh Gottheimer and Mike Lawler, would direct the National Institute of Standards and Technology to write security standards for AI agents, covering verification of agent actions, tamper-proof activity logs and machine-readable inventories of deployed agents, according to a summary by Paubox. The superintelligence ban goes further than any of that. It does not ask how the most powerful systems should be regulated. It asks whether they should be built at all.