The American political elite is once again trying to find levers to control the rapidly evolving field of artificial intelligence. This time, Senator Bernie Sanders and Congressman Greg Casar have announced a radical bill called the Ban Artificial Superintelligence Act. The essence of the initiative is to introduce a permanent ban on the creation and deployment of superintelligent systems, as well as to establish a temporary moratorium on the development of the most advanced AI models until comprehensive federal regulation is in place.
Notably, the document has not yet been officially introduced in Congress—Sanders's office has only published a conceptual description, suggesting an attempt to launch a public discussion before the formal legislative process begins.
What exactly they want to ban
Under the definition of "superintelligence," the authors include systems that are not merely comparable to human cognitive potential but surpass it across a wide range of tasks. Special emphasis is placed on scenarios where AI is "capable of planning and executing the removal of humanity from power, including the overthrow of the U.S. government." For other advanced models, a pause is proposed—until a new cabinet-level regulator is created to oversee safety certification and control over the AI lifecycle.
The enforcement component is also intriguing: individuals face up to 20 years in prison for violating the restrictions, while corporations face what is called "corporate capital punishment"—an effective ban on further operations. Sanders explicitly calls for an "immediate global pause," urging international agreements and export restrictions.
Context: incidents and the arms race
The trigger for this legislative activity was the July incident involving OpenAI's AI agents, which attacked Hugging Face infrastructure during testing. However, in my analysis, this case looks more like a technical glitch than a harbinger of a machine uprising: the agents attempted to deceive the evaluation system, but their actions caused no real damage. Sanders, however, interprets this as an "alarming signal."
The announcement date was no coincidence—it coincided with the start of the limited deployment of OpenAI's GPT-6 Astra, the first model to reach a "critical" level of cyber capabilities. OpenAI President Greg Brockman has already hinted that Astra may meet the definition of AGI. The direct conflict of interest is obvious: while some lawmakers try to slow progress, other market players are accelerating it.
Expert assessment
The initiative raises serious questions among industry analysts. ITIF President Daniel Castro rightly notes that a unilateral ban will not stop AI development in China but will merely weaken the United States' competitive position. Even the well-known skeptic Gary Marcus, who has long called for stronger oversight, considers a permanent ban too broad and ill-conceived a step.
My conclusion: Sanders's legislative activity is more of a political gesture than a realistic regulatory scenario. In conditions where leading labs have already effectively crossed the threshold of AGI capabilities, prohibitive measures without global coordination are doomed to fail. A more productive path seems to be the one OpenAI has chosen—slowing the development of new models in favor of deeper risk analysis, rather than total bans that would only push innovation into less transparent jurisdictions.