Senator Bernie Sanders and Representative Greg Casar introduced legislation on September 23 that would establish a permanent ban on artificial superintelligence development and deployment across the United States. The bill marks the first major congressional effort to preemptively restrict an AI capability tier that does not yet exist in practical form.
The proposal addresses what researchers call artificial superintelligence (ASI), a theoretical system with intellectual capabilities exceeding human performance across all domains. No such system exists today. Current large language models and AI systems operate within narrowly defined tasks, even when they perform those tasks at superhuman levels. The bill treats superintelligence as a distinct regulatory category requiring complete prohibition rather than managed development.
The legislation pairs the superintelligence ban with creation of a new federal artificial intelligence agency. This agency would house regulatory authority over AI development across the country. The structure reflects growing congressional recognition that AI governance requires dedicated institutional capacity. Existing regulatory bodies lack specialized expertise and jurisdiction spanning the fragmented AI landscape.
Sanders and Casar's approach diverges sharply from the regulatory consensus emerging in other democracies. The European Union pursued the AI Act, which classifies AI systems by risk level and imposes graduated restrictions. This risk-based framework allows high-capability systems to operate under strict oversight. The United Kingdom adopted lighter regulatory guidance designed to support AI innovation. Canada recently implemented rules for high-impact AI systems but stopped short of banning capability categories.
The permanent ban language creates enforceability challenges. Definition becomes paramount. Superintelligence lacks a universally accepted technical definition. Lawmakers must determine whether systems meeting certain benchmark performance thresholds trigger the ban or whether the restriction targets only systems meeting speculative theoretical criteria. Poor definition invites either regulatory capture by industry players narrowing the scope or overreach that inadvertently restricts conventional AI development.
The bill's reception splits predictably. AI safety advocates including some effective altruism researchers support preemptive bans on hypothetical advanced systems, arguing that preventive regulation beats emergency measures after harm occurs. Industry groups and many AI researchers oppose blanket bans, contending that undefined superintelligence categories freeze beneficial research and that graduated oversight suits the technology better. Some argue the U.S. cannot unilaterally ban superintelligence when AI labs in China, Europe, and other nations continue advancing capability development.
Practical implementation faces obstacles. Congressional gridlock around AI regulation persists despite bipartisan agreement that action is necessary. The distinction between regulation and prohibition matters enormously. A regulatory agency can adjust rules as technology evolves. A permanent ban cannot easily accommodate new information about superintelligence feasibility or timeline. If superintelligence emerges as an achievable target within 20 years rather than 100, the ban becomes live policy. If it remains theoretical, the ban creates regulatory theater without practical effect.
The bill enters a crowded legislative landscape. Multiple committees claim AI jurisdiction. Senate Democrats, Senate Republicans, and House members have proposed conflicting frameworks. The Commerce Committee, Energy Committee, and Intelligence Committee compete for oversight authority. Without consensus on whether superintelligence merits unique treatment or fits within broader AI governance, passage faces long odds.
Sanders and Casar's initiative signals that superintelligence governance has moved from academic thought experiment to congressional agenda. Whether the permanent ban approach survives committee review and floor votes remains uncertain. The outcome will shape how American law treats theoretical AI capabilities for years ahead.