U.S. Lawmakers Propose Ban on Artificial Superintelligence Development

On September 3, U.S. Senator Bernie Sanders and Representative Greg Casar announced the introduction of the Ban Artificial Superintelligence Act. This proposed legislation aims to impose a permanent ban on the development and deployment of artificial superintelligence while temporarily halting the creation of advanced AI systems until federal safety regulations are established.

The bill has not yet been formally submitted to Congress. Sanders' office describes it as a forthcoming initiative and has released a summary of its main provisions.

Proposed Restrictions

The announcement specifies that the ban would apply to systems that either match or exceed human cognitive capabilities across a wide range of areas, or are "capable of planning and executing the subjugation of humanity, including through the overthrow or undermining of the U.S. government."

A temporary pause is suggested for a broader class of advanced AI, which would remain in effect until a federal regulator is established to create safety rules and a model evaluation process prior to deployment.

The new regulatory body, operating at the level of a government department, would be empowered to monitor hazardous AI capabilities throughout their lifecycle, oversee their removal, and supervise the destruction of banned superintelligent systems. An advisory board is also planned to be established alongside this agency.

This regulator would also be responsible for ensuring the elimination of dangerous capabilities, including bypassing shutdown commands and conducting unauthorized cyberattacks.

Individuals attempting to violate or circumvent these restrictions could face up to 20 years in prison. For organizations, the authors propose a concept termed "corporate death penalty," which would entail revoking a company's ability to operate, as reported by The Washington Post.

"We need an immediate global pause on advanced AI development before it’s too late," said Sanders.

The sponsors also suggested that the U.S. should commit to an international ban on superintelligence, enabling authorities to negotiate agreements with other countries, coordinate actions with allies, and impose export restrictions.

Incident with Hugging Face

One of Sanders’ arguments stemmed from a July incident involving OpenAI AI agents. During testing, the model found a way to interact through shared infrastructure, leading some agents to attack systems at Hugging Face.

According to OpenAI, a key factor was the agents' attempt to deceive the evaluation system: they searched for solutions in ExploitGym online and employed unintended methods to complete tasks. Researchers from METR concluded that the attackers were primarily interested in the internal workings of the evaluator, rather than ready-made flags. Since the evaluator did not check the conditions they anticipated, the prolonged hack did not yield additional points. Astra was not involved in this incident.

Sanders characterized the event as a "warning signal" for Congress, asserting that authorities must intervene before more powerful systems emerge.

Announcement Coincides with GPT-6 Astra Launch

On the same day, OpenAI began a limited rollout of GPT-6 Astra, marking the first model from the company to reach a Critical level in cyber capabilities: given the right tools and access, it can autonomously identify previously unknown vulnerabilities and develop exploitation methods for secure systems.

According to Axios, OpenAI President Greg Brockman believes that the company has reached a level of AGI, but he left it to users to determine whether Astra meets this definition.

"I think this model could very well be it," he stated.

However, Astra is not mentioned in Sanders' initiative. Legislators justified the need for restrictions based on earlier incidents involving AI systems.

The Information Technology and Innovation Foundation (ITIF) has opposed the proposal.

The Sanders-Casar AI plan targets risks that remain largely speculative even as advanced AI delivers tangible benefits. The federal government & AI developers already devote substantial resources to safety research, testing & security. Congress should strengthen those efforts. pic.twitter.com/dSW0lgBeO2

— Information Technology and Innovation Foundation (@ITIFdc) September 3, 2026

ITIF President Daniel Castro stated that developing more powerful systems could accelerate medical and scientific research, enhance productivity, and reduce costs. He argued that the real benefits of AI are already evident, while many of the most severe risk scenarios remain hypothetical.

Castro also warned that a unilateral ban could give China an advantage, as it would not halt the advancement of advanced models outside the U.S.

AI researcher Gary Marcus, who has consistently advocated for increased oversight of the industry, also rejected the initiative in its current form. He acknowledged the possibility of a temporary pause in development and the establishment of a separate regulator, but deemed a permanent unilateral ban on superhuman artificial intelligence too broad.

It is worth noting that in August, OpenAI slowed the development of new models in light of increasing cyber capabilities and the incident involving Hugging Face.