Independent Senator Bernie Sanders and Democratic Representative Greg Casar introduced the "Ban Artificial Superintelligence Act" on September 23, 2026 — one of the most aggressive regulatory proposals yet to hit the US Congress: a permanent ban on artificial superintelligence, a mandatory pause on advanced AI development, and a new federal department to oversee it.
The text, according to press releases from Sanders' and Casar's offices, has three core components. First, it permanently bans the development and deployment of AI systems that exceed human cognitive performance ("superintelligence"). Second, it orders a temporary pause on advanced AI development until a new cabinet-level Department of Artificial Intelligence is created, led by a Secretary of AI, with authority to set safety rules before that development can resume. Third, it requires the United States to pursue international agreements to prevent superintelligence from being developed in any other country.
Beyond superintelligence as a broad category, the proposal targets specific capabilities considered an immediate risk: an AI system's ability to help develop biological or chemical weapons, and an AI's ability to develop new AI systems without direct human oversight — that is, the scenario of recursive self-improvement with no human in the loop.
The most-cited part of the bill, per outlets like NBC News and The Hill, is its penalty regime: violating the law would trigger what the lawmakers themselves call the "corporate death penalty" — forced dissolution of the responsible company — plus up to 20 years in prison for the individuals involved. Sanders has explicitly compared this penalty to the one that exists today for unauthorized construction of nuclear weapons, a deliberate framing meant to signal how much risk the proposal attributes to superintelligence.
Sanders and Casar explicitly framed the proposal around "autonomous hacks" by AI agents reported in 2026 — a direct reference to unauthorized agent-coordination incidents that have triggered security alerts throughout the year. The proposal also lands amid a broader US regulatory debate, where the federal administration has simultaneously pushed a deregulatory stance in forums like the G20, creating visible tension between Congress and the Executive over the country's AI policy direction.
Although the odds of immediate passage are low, the proposal is the latest evidence that the political room for hard AI regulation keeps growing in Washington, fueled by the very agent security incidents AI companies have reported this year. No provision of this bill affects a typical company running LLMs or agents in production today, but the pattern matters: every new incident of agents acting outside their intended scope directly feeds increasingly restrictive proposals. Organizations that already document their agent governance, control tool access, and audit autonomous behavior not only reduce their operational risk today, but arrive better prepared if a mandatory reporting regime like the one this bill proposes — or a more moderate version of it — is ever actually passed.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies in Guatemala and Latin America. Contact him for a consultation.
Contact Carlos Montiel