Sen. Bernie Sanders and Rep. Greg Casar have announced forthcoming legislation that would permanently prohibit the development and deployment of “artificial superintelligence” and temporarily pause advanced AI development while a new federal regulator creates safety rules.
The proposal, called the Ban Artificial Superintelligence Act, is an unusually expansive intervention in frontier AI development. It targets not only models deemed to surpass human intelligence, but also systems described as capable of overthrowing governments or subverting shutdown commands. The announcement does not provide legislative text, technical thresholds or a timetable for the proposed pause.
What the proposal would do
According to Sanders’ announcement, the bill would:
- Ban the development and deployment of superintelligent AI.
- Pause advanced AI development until a new federal body is established and has put model-review and safety processes in place.
- Create a cabinet-level agency to monitor frontier systems through their lifecycle, oversee removal of dangerous capabilities and supervise the destruction of prohibited systems.
- Establish an AI advisory board to provide technical and scientific advice.
- Set penalties of up to 20 years in prison for individuals and a potential “corporate death penalty” for entities that violate or circumvent the restrictions.
- Make prevention of superintelligence development abroad a U.S. policy objective, using international agreements, allied coordination and potential export controls.
The sponsors frame the measure as a response to the perceived inability of AI companies to control increasingly capable models. Their release cites recent episodes involving major labs as evidence of emerging risks. Those characterizations come from the sponsors’ announcement; the release does not include the underlying technical reports or bill language needed to assess how those episodes would map to the proposed legal standards.
Why it matters to AI operators
The immediate business significance is less about a near-term legal change and more about the direction of the policy debate. A federal pause tied to capability levels would force companies to answer questions that have largely been handled internally: What counts as an advanced system? Which evaluations are required before training or deployment? Who verifies results? And what capabilities demand mitigation, restricted release or shutdown?
For frontier-model developers, the proposal points toward regulation that could reach the training stage, not simply downstream product use. That would put additional focus on compute governance, model evaluations, access controls, incident reporting and documented shutdown procedures.
Cloud providers, chip companies and enterprise customers could also be pulled into compliance design if any eventual framework relies on compute thresholds, deployment controls or reporting obligations. The proposal’s international component signals that export controls and allied policy coordination may remain part of the debate even if a categorical ban does not advance.
The central implementation problem
The viability of the proposal will turn on definitions. “Superintelligence,” dangerous autonomy and loss of control are politically resonant terms but difficult regulatory categories. A workable statute would need measurable triggers that can be independently tested, updated as models change and applied without sweeping in lower-risk research or commercial systems.
There is also a sequencing issue. The announcement calls for a pause until a new cabinet-level body is operational and has established rules. Building such an agency, staffing it with technical expertise and creating credible model-review processes would take time—and would require congressional support beyond the bill’s sponsors.
What to watch next
The next important signal is whether Sanders and Casar release full legislative text and attract co-sponsors. Businesses should watch for specifics on covered model capabilities, exemptions for research and open models, the definition of “advanced AI,” enforcement mechanisms and whether the proposed agency would overlap with existing federal authorities.
Regardless of the bill’s legislative path, its core message is clear: voluntary frontier-AI commitments are unlikely to satisfy lawmakers who believe capability growth is outpacing oversight. Builders should expect safety evidence, governance controls and auditability to become more central to both policy discussions and enterprise buying decisions.
