Sen. Bernie Sanders and Rep. Greg Casar are advancing a proposal that would permanently ban certain artificial-superintelligence systems, temporarily pause advanced AI development and create a cabinet-level federal body to regulate the technology. The measure, described by its sponsors in early September and reported by the Associated Press as being unveiled Sept. 23, would put federal approval between developers and deployment of advanced systems once new safety rules are in place.
The proposal is notable less for a claim that present-day chatbots have crossed a known technical line than for the regulatory structure it sketches: an outright prohibition for a defined class of system, an interim development pause, pre-deployment oversight, criminal penalties and an international strategy aimed at stopping prohibited systems elsewhere. Its legislative path is uncertain. Associated Press reporting carried by ABC News says the proposal faces long odds in the Republican-controlled Congress.
A ban paired with a temporary pause
Sanders and Casar first announced their forthcoming legislation on Sept. 3 in a statement from Sanders’ Senate office. It described two different restrictions. Development and deployment of artificial superintelligence would be permanently barred, while advanced AI development would be paused temporarily until the federal government established safety rules.
That split is important because the proposal does not appear to treat all sophisticated software as superintelligence. AP reported that its definition covers a system that exceeds human cognitive ability, or one with enough capability to plan and execute humanity’s destruction or disempowerment. The public descriptions do not set out a measurable technical threshold, such as computing capacity, benchmark scores or training cost, for determining when a system enters the regulated category.
The distinction also leaves an unresolved practical question: how regulators would assess systems that are powerful in narrow areas but lack broad autonomy, or that exhibit high benchmark performance without being able to reliably act in the world. Those details normally reside in statutory language and implementing rules. Neither the sponsors’ announcement nor the AP account identified a bill number or linked filed legislative text, so the proposal’s formal definitions and exemptions cannot yet be independently checked from those public accounts.
A proposed federal AI department
The sponsors say their measure would establish a new cabinet-level agency, advised by an Artificial Intelligence Advisory Board. AP calls the proposed body the Department of Artificial Intelligence; the Sept. 3 announcement does not provide a formal agency name. Both descriptions point to a regulator with unusually broad powers over frontier systems.
According to the sponsors’ outline, the agency would monitor frontier AI for dangerous capabilities, supervise the removal of those capabilities and oversee the destruction of artificial superintelligence. AP reported that, after the agency creates safety rules, advanced AI systems would require federal approval before deployment.
In operational terms, that would create a sequence rather than a single rule: the temporary pause would apply while standards are being developed; developers would then need approval to deploy advanced systems; and systems meeting the superintelligence definition would remain prohibited. The public material does not establish how the agency would distinguish an advanced model eligible for review from a prohibited one, how it would test models, or whether restrictions would apply to model weights, training runs, deployment access or all of those stages.
Penalties described by the sponsors are also severe. Their announcement says organizations that violate the law could face a corporate death penalty and individuals could face prison terms of up to 20 years. AP likewise reported that some violations could carry terms of up to 20 years. The announcement does not explain what conduct would trigger each sanction or what “corporate death penalty” would mean as a statutory remedy.
International controls are part of the design
The proposal extends beyond domestic developers. The sponsors say it would direct the United States to seek international agreements and coordinate with allies to prevent artificial superintelligence worldwide. It also contemplates export controls, placing the proposal in a policy area where software governance intersects with control over advanced computing equipment and cross-border technical cooperation.
That international component recognizes a basic enforcement problem: a U.S. prohibition would not by itself prevent work in another jurisdiction. But agreements, allied coordination and export controls are policy tools with different reach and enforcement burdens. The available descriptions do not say which countries or technologies would be covered, or how the government would determine whether a foreign system met the bill’s definition.
Supporters frame the approach as a response to catastrophic risks from systems that could become uncontrollable. Several employees at leading AI companies endorsed the legislation in a statement shared with AP, according to its report. Those views are political and technical judgments, not evidence that a currently deployed system meets the proposed legal definition. The sponsors’ Sept. 3 statement also makes broad assertions about AI behavior at major companies, but the public accounts cited here do not independently substantiate those assertions.
A September rollout, with details still to come
The proposal’s public rollout unfolded over several weeks. On Sept. 3, Sanders and Casar said they would introduce the measure. In prepared remarks published Sept. 15, Sanders said he expected to introduce it with Casar the following week. AP then reported Sept. 23 that the lawmakers were unveiling the legislation after providing it to the news agency.
There is also a naming inconsistency in the sponsors’ own Sept. 3 announcement: its headline uses “Ban Artificial Superintelligence Act,” while the body refers to an “Artificial Superintelligence Ban Act.” That may be a drafting variation, but a formal title, bill number and text would settle it.
For now, the proposal is best understood as a detailed demand for a new federal gatekeeping system around the most capable AI, rather than a law that has taken effect or an agency that exists. Whether Congress takes it up will determine whether its definitions, approval process and enforcement powers receive the scrutiny that the brief public outline leaves open.