U.S. Senator Bernie Sanders and Representative Greg Casar Officially Submit the "Artificial Superintelligence Prohibition Act"
U.S. Senator Bernie Sanders and Representative Greg Casar have officially submitted the "Artificial Superintelligence Prohibition Act," which permanently bans the development and deployment of superintelligence, imposing a maximum penalty of 20 years in prison for violators.
The official definition states: systems that exceed human cognitive performance in most areas or possess the capability to destroy or deprive human power, including the ability to overthrow the federal government. The prohibition covers development, internal or external deployment, acquisition, possession, funding, importation, and transfer, as well as systems exhibiting precursors to superintelligence. Upon discovery of precursors, systems must be isolated from the internet and deactivated within 30 days unless the developer can remove these features. Advanced artificial intelligence must be suspended until a newly established cabinet-level AI department formulates testing, reporting, auditing, and pre-launch approval rules; unlaunched advanced systems cannot be deployed during this period.
Penalties are written in the style of illegal nuclear development: decision-makers or "rogue actors" face up to 20 years in prison; non-decision-making individuals may be banned from working in the AI industry for ten years; companies may have their charters revoked, and their intellectual property and assets transferred to the federal government for the destruction of related systems and hardware, referred to in the proposal as "corporate death penalty." A new department will be assisted by a scientific advisory committee to monitor the entire lifecycle of frontier systems, oversee the removal of dangerous capabilities, and ensure the destruction of superintelligence. Foreign policy will seek international agreements, ally collaboration, and export controls to prevent superintelligence from being developed anywhere. Sanders stated that rushing towards a cliff cannot be solved by merely easing off the gas; Casar remarked that regulating cutting-edge technology is less effective than regulating ordinary food trucks.
Elon Musk retweeted discussions related to this and only responded with a thinking emoji. The office of President Donald Trump has requested official documents to refer to artificial intelligence as superintelligence and opposes global uniform tightening. The bill's prospects remain unclear, but it directly targets individual criminal liability in training and deployment, aimed at laboratory leaders and funders.
In market mechanisms, this represents regulatory shock rather than an immediate ban. Buyers may temporarily increase investments in companies with a "compliance narrative"; sellers are closed-source laboratories and cloud computing contracts that are pushing for next-generation training. Funding may delay large training commitments, shifting towards products that can be defined as non-cutting-edge models. Beneficiaries will be policy alliances advocating for evaluation licenses, while those under pressure are U.S. laboratories expanding parameters and computing power at maximum speed; if the U.S. unilaterally pauses while other countries do not, computing power and talent will overflow.
On a supplementary level, the text includes "easily convertible to superhuman" in its definition, expanding the enforcement interpretation space. There is no consensus in the scientific community on the operational definition of superintelligence. The proposal also requires that the development of advanced systems must first obtain permission from the new department.
Source: Public Information
ABAB AI Insight
Sanders frames artificial intelligence as nuclear material: the prohibition is not on applications but on capability thresholds. The corporate death penalty and 20-year prison sentences shift the training decisions of limited liability companies to personal criminal risks, altering the logic of board signatures—general counsels will require any training approaching the threshold to be halted. The establishment of a new cabinet department equates to creating a separate licensing authority beyond the existing Departments of Commerce, Energy, and Defense, with a scientific advisory committee aimed at transferring the definition of "who counts as superintelligent" from laboratory public relations to administrative interpretation.
Capital pathways will first avoid criminal liability and then avoid technology. If U.S. funds and cloud providers find the definitions ambiguous, they will relocate large training to jurisdictions with looser interpretations or break projects into "non-advanced" slices to bypass the pause. The export control clauses attempt to block this route, at the cost of tightening computing power politics with China and allies. Musk's response of a thinking emoji neither endorses the ban nor supports the laboratories, leaving space to treat superintelligence as both a product name and a risk name.
Similar structures exist in the Atomic Energy Act, the Biological Weapons Convention, and some states' permanent bans on cloning: first using existential threats to exchange for criminal clauses, then deciding in the definition battle whether the industry can breathe. The industry is at a stage where laboratories admit insufficient control, while the White House seeks to compete for leadership. The UN Security Council has just heard Altman's "extreme caution," while progressive factions in Congress propose a permanent ban and corporate death penalty, with the two sets of language not on the same negotiation table.
The essence is regulatory change. The mechanism is: when capabilities are written as potentially overthrowing the government, training shifts from commercial activity to quasi-weapon behavior; once criminal liability falls on natural persons, pauses do not need to go through budgets, as long as prosecutors are willing to interpret "precursor features." The real limitation does not come from whether the bill can pass today, but from insurance companies, boards, and cloud contracts starting to quote based on the worst definitions.
ABAB News · Cognitive Law
- What is prohibited is the threshold; the power of interpretation is more lethal than the text.
- Companies can go bankrupt, but signatories cannot escape prison terms.
- If one lane brakes unilaterally while the adjacent lane accelerates, the vehicle will cross the border.