Democratic lawmakers are renewing their effort to restrict advanced artificial intelligence, with Representative Yassamin Ansari calling for a development pause and a prohibition on artificial superintelligence following the party's midterm election gains. The proposal would place some of the most powerful AI systems under federal control and potentially make their development illegal.
Ansari made her position public on November 4, bringing renewed attention to legislation that would impose substantial limits on frontier AI research. The proposals go beyond conventional industry regulation: they would establish a dedicated federal department, suspend certain advanced AI development activities and prohibit artificial superintelligence altogether. According to The 420.in, however, the measures remain pending before Congress rather than representing enacted government policy.
Democrats renew AI push
Ansari wants Democrats to make three priorities part of their political agenda: stopping advanced AI development temporarily, introducing tougher industry oversight and preventing artificial superintelligence from being created. Her intervention comes after Democratic gains in the midterm elections, giving the proposals renewed political visibility.
The legislative effort predates her latest comments. Two bills introduced in September seek to establish restrictions on highly capable AI technologies, including systems that lawmakers believe could develop capabilities beyond meaningful human control. Ansari is among the sponsors of the House legislation.
The renewed attention matters because the proposals address a fundamental question in AI governance: whether governments should regulate dangerous applications of powerful AI or prohibit the development of certain capabilities before they emerge. The legislation adopts the latter approach for artificial superintelligence.
Two bills target superintelligence
The House version, H.R. 10538, carries the title Ban Artificial Superintelligence Act of 2026. Representative Greg Casar introduced it on September 24, with Ansari listed among its co-sponsors.
A corresponding Senate measure, S. 5493, was introduced a day earlier, on September 23, by Senator Bernie Sanders. Together, the proposals seek to create a federal framework that would govern the development, deployment and oversight of advanced AI technologies.
Both measures envision a new federal Department of Artificial Intelligence. The proposed agency would oversee safety requirements and determine when advanced systems could be developed or released. Until the department was fully established and its safety rules were operational, the legislation would require a mandatory suspension of covered AI development.
Computing power defines AI
Rather than relying exclusively on a system's demonstrated abilities, the legislation would primarily classify advanced AI according to the computational resources used during training. Its proposed threshold is 10²⁵ operations, counting integer and floating-point calculations.
The threshold would not necessarily remain fixed. The legislation includes provisions allowing adjustments as improvements in training techniques make computing resources more efficient. This approach attempts to account for technological progress rather than treating one computing benchmark as permanently sufficient to identify advanced systems.
The proposed suspension would cover more than the initial training of a model. Developers would also be prevented from modifying or fine-tuning covered systems during the pause. Fine-tuning means taking an existing trained model and adapting it for particular tasks or behaviours.
There would be limited exceptions. One identified purpose would be removing capabilities regarded as potential precursors to superintelligence. This distinction is significant because the proposed restrictions would not treat every modification as equally dangerous: certain work intended to reduce risks could still be permitted.
Deployment would need approval
The restrictions would also affect systems that have already been developed but have not yet been released. Under the proposals, developers could not deploy unreleased advanced AI systems until the new federal department had established requirements for safe development and operation.
Those requirements would cover monitoring, safety audits and government approval before deployment. Developers would eventually need a federal charter to work on advanced AI models, while releasing a covered system would require authorisation from the department.
The proposed department would also receive substantial inspection powers. It could examine advanced AI systems and determine whether their capabilities included characteristics prohibited by the legislation. If officials identified dangerous precursor characteristics, they could order a system disconnected from the internet.
Where the identified risks could not be eliminated, the department could potentially require the system to be made inoperative. This would give the proposed agency authority not only to establish rules but also to intervene directly when a system was considered unsafe.
Superintelligence would be prohibited
The legislation's most far-reaching provision is its proposed prohibition on artificial superintelligence. It would cover a wide range of activities, including developing, deploying, acquiring, possessing, financing, importing and transferring prohibited systems.
The bill defines artificial superintelligence through two broad criteria. One concerns AI that performs beyond human cognitive abilities across most fields. The other covers systems capable of planning and carrying out actions intended to destroy humanity or strip people of their power.
The proposal also identifies capabilities that lawmakers consider possible warning signs of future superintelligence. These include systems that can accelerate AI research, obtain unauthorised access to infrastructure, resist shutdown attempts or improve weapons development.
Other listed characteristics involve the ability to change a system's own functions and evade human supervision. The legislation therefore focuses not just on a system's present intelligence but also on behaviours that could make it increasingly difficult for people to control.
The proposed ban would extend to systems displaying specified precursor characteristics, not just AI already classified as superintelligent. That distinction could make the restrictions relevant well before a system meets the bill's full definition of artificial superintelligence.
Penalties would be severe
The bills also propose substantial consequences for violations. Policymakers and individuals described as rogue actors who recklessly breach key provisions could face financial penalties and prison sentences of up to 20 years.
Other individuals who recklessly violate the relevant requirements could face a different professional restriction. Those outside policymaking positions could be prohibited from working in the AI industry for 10 years, although the proposal provides for appeals and certain limited exceptions.
These provisions indicate that the proposed framework would reach beyond corporate compliance departments and government regulators. Depending on the circumstances, individuals involved in prohibited AI activities could face personal legal and professional consequences.
The new department's inspection and enforcement authority would operate alongside these penalties. Its ability to order internet isolation or potentially disable systems would provide an additional mechanism for responding to identified risks.
AI safety debate remains divided
The proposals arrive amid disagreement over which risks from advanced AI deserve the greatest attention. Some technology executives and researchers have warned that increasingly sophisticated systems could eventually become difficult for humans to control.
Other observers place greater emphasis on immediate problems arising from the use of existing technologies. These include malicious human activity, excessive corporate concentration, misinformation, cybercrime and inadequate governance.
The distinction matters. One approach seeks to prevent dangerous uses and strengthen oversight of companies developing AI. The other, reflected in the proposed legislation, treats the creation of superintelligence itself as a potential threat to public safety.
"a ban on artificial superintelligence"
— Yassamin Ansari, US congresswoman
Ansari's November 4 intervention brings this more restrictive position back into the political discussion. It also highlights growing pressure for governments to establish legal boundaries before frontier AI systems acquire substantially greater capabilities.
Congress has not approved bans
Despite the renewed political attention, both proposals remain at an early stage. The House bill was referred to the House Committee on Oversight and Government Reform, while its Senate counterpart went to the Senate Committee on Commerce, Science and Transportation.
Neither measure has passed its respective chamber. That means the proposed restrictions have not become federal law, and current AI models are not subject to an immediate nationwide development prohibition under these bills.
The next important developments will come from the congressional process. Committee consideration, further legislative action and the positions taken by lawmakers will determine whether the proposals move forward, change substantially or remain stalled.
It is also possible that congressional debate could focus on a less restrictive regulatory framework rather than adopting the full proposed pause and prohibition. The legislation presents several distinct policy choices, from temporary development restrictions to permanent prohibitions on specified AI capabilities.
Why it matters
The renewed Democratic push places artificial superintelligence at the centre of a legislative debate that could shape the future of US AI development. Instead of relying solely on voluntary industry safeguards or rules governing individual applications, the proposals would give the federal government authority over which advanced systems could be developed and released.
For AI companies and researchers, the distinction is substantial. The proposed framework could affect model training, experimentation, funding, deployment and professional participation in the industry. It would also create a government approval process and potentially criminal consequences for prohibited conduct.
For now, these are proposed restrictions, not operative requirements. The immediate question is whether Ansari's renewed call translates into greater congressional support for the bills introduced by Casar and Sanders. Their progress through the two committees will provide the clearest indication of whether the United States moves toward a mandatory pause and a prohibition on artificial superintelligence, or chooses a narrower system of AI oversight instead.