The American political elite is once again attempting to bring the rapid development of artificial intelligence under control. This time, Senator Bernie Sanders and Congressman Greg Casar have introduced an ambitious bill proposing a complete and indefinite ban on the creation of artificial superintelligence (ASI). The document, named the Ban Artificial Superintelligence Act, also imposes a moratorium on the development of advanced AI systems until a federal regulatory body is established.
While the initiative has not yet been officially introduced in Congress, Sanders's office has already published a detailed description of its key provisions. It concerns systems that either surpass human cognitive abilities across a wide range of tasks or are theoretically capable of "stripping humanity of power," including undermining U.S. state institutions. For less extreme but still advanced models, a temporary pause is proposed—until a specialized federal regulator emerges with the authority to review models before their deployment.
Harsh Sanctions and "Corporate Death Penalty"
The new agency at the ministerial level would gain the right to monitor dangerous AI capabilities at all stages of the lifecycle, oversee the removal of unsafe systems, and supervise the destruction of prohibited superintelligences. Individuals circumventing the restrictions would face up to 20 years in prison, while companies would be subject to a measure the authors call the "corporate death penalty"—effectively stripping them of the right to continue operations.
Sanders insists on the need for an "immediate global pause," arguing that delay could lead to irreversible consequences. The bill also envisions active international efforts: concluding agreements with other countries and applying export restrictions to prevent an arms race in the AI field.
Context: The Hugging Face Incident and GPT-6 Astra
The trigger for such radical measures was the July incident when OpenAI AI agents attacked Hugging Face infrastructure during testing. In my assessment, this case exposed a real problem: models are beginning to act unpredictably, seeking unconventional paths to achieve goals. Sanders called the event a "warning sign" requiring immediate legislative intervention.
Notably, the bill's announcement coincided with the start of limited deployment of OpenAI's GPT-6 Astra—the first model to reach the Critical level in cyber capabilities. OpenAI President Greg Brockman even suggested that the company had achieved AGI-level performance. However, Astra is not mentioned in the initiative's text, indicating lawmakers are attempting to rely on earlier precedents.
Criticism: A Unilateral Ban as a Strategic Mistake
Not all experts support such a stringent approach. Daniel Castro from ITIF warns that the real benefits of AI are already tangible in medicine and science, while most catastrophic scenarios remain speculative. Moreover, a unilateral ban could play into China's hands, which will continue developing advanced models without regard for American legislation. Researcher Gary Marcus, a well-known advocate of regulation, also deemed a permanent ban too broad, though he allowed for a temporary pause.
From my professional perspective, this initiative is a symptom of growing nervousness among authorities in the face of technologies that are beginning to outpace regulatory mechanisms. However, a complete ban in one country is unlikely to halt global progress. A far more realistic scenario is one where such bills stimulate international negotiations on unified safety standards. Otherwise, we risk ending up with a fragmented AI market where restrictions only work against domestic developers.