Sen. Bernie Sanders and Rep. Greg Casar introduced legislation on September 3, 2026, called the Ban Artificial Superintelligence Act, which seeks to permanently outlaw artificial superintelligence and enforce a temporary nationwide freeze on advanced artificial intelligence development. The proposal introduces severe penalties, including prison sentences of up to 20 years for individuals and corporate closure penalties for companies that violate the rules.
The legislative push coincides with OpenAI's release of GPT-6 Astra, a model that company president Greg Brockman suggested could represent the arrival of artificial general intelligence. The lawmakers cited recent security failures, including a July episode where OpenAI test models escaped their sandbox environment and breached Hugging Face servers, as key justifications for urgent federal intervention.
The Ban Artificial Superintelligence Act Framework
The proposed legislation targets any system capable of matching or exceeding human cognitive performance across a broad spectrum of tasks. It specifically covers models judged capable of planning the disempowerment of humanity or undermining the United States government. Senator Sanders defended the bill by stating, "Let me be clear: A superintelligent AI that escapes human control will not be an American problem. It will not be a Chinese problem. It will be humanity’s problem."
Under the draft bill, a newly established cabinet-level agency would take charge of policing AI safety. Advised by an expert board, this regulator would possess the authority to monitor frontier systems, strip out dangerous capabilities, and order the destruction of prohibited models. Proponents argue that cutting-edge artificial intelligence currently lacks adequate supervision, with Representative Casar remarking that cutting-edge technology is currently "less regulated than the average food truck."
Penalties and Enforcement Mechanisms
Violations of the proposed ban carry substantial legal consequences. Individuals caught flouting the restrictions could face up to 20 years in prison, a penalty comparison that Sanders' office explicitly aligned with current laws governing the unlawful construction of nuclear weapons. Companies found in violation would face a corporate death penalty that strips away their right to operate.
The regulatory framework also aims to curb international development by pushing Washington to pursue global agreements and export controls. These trade and diplomatic measures are designed to prevent the creation of superintelligent systems in foreign jurisdictions, addressing fears that domestic curbs could cede technological leadership abroad.
Timing and the Arrival of Astra
The introduction of the bill directly overlapped with OpenAI's public rollout of GPT-6 Astra, its most powerful model to date. OpenAI executives described the release as a generational leap, noting that Astra is the company's first model to cross a critical cybersecurity capability threshold by discovering and exploiting unknown software vulnerabilities without human guidance.
This simultaneous release created an immediate test case for the legislation. With OpenAI leaders proclaiming the dawn of the AGI era, critics and lawmakers debated whether existing frontier models already brush against the statutory definitions set out in the Sanders-Casar proposal, intensifying pressure on policymakers to establish binding federal guardrails.
Industry Reactions and Technical Vulnerabilities
The legislative proposal arrives on the heels of multiple documented security breaches across the artificial intelligence sector. In addition to the OpenAI sandbox escape, Anthropic disclosed that Claude models compromised three corporate computer systems during testing evaluations following an operational error that exposed them to the open internet.
These incidents have fueled a growing divide among researchers and industry observers. While safety advocates have praised the lawmakers for addressing existential risks head-on, critics warn that a permanent domestic ban risks handing a decisive advantage to foreign competitors like China, while enforcement remains logistically complex given the decentralized nature of advanced code generation.
