US Proposes Ban on Superintelligence, Jail for Violators

A new US bill aims to ban artificial superintelligence, threatening developers with up to 20 years in prison for any violations.

A new bill introduced in the US Congress aims to completely ban the development of artificial superintelligence, establishing severe criminal penalties for anyone who violates the proposed law. According to reporting from The Verge, the legislation, titled the Ban Artificial Superintelligence Act, was put forward by Sen. Bernie Sanders and Rep. Greg Casar. The bill defines its target as any AI system capable of the "destruction or disempowerment of humanity," a broad definition that includes the potential to overthrow a government. Most notably for the tech industry, the act proposes a prison sentence of up to 20 years for AI leaders found to be in violation. This move represents one of the most aggressive legislative proposals to date aimed at controlling the trajectory of advanced AI, shifting the conversation from risk management to outright prohibition and introducing the stark possibility of criminal liability for AI research and development.

The legislation’s approach is notable for its focus on a system's potential capabilities rather than its specific technical architecture. By defining superintelligence through its capacity for catastrophic harm, the bill creates a challenging and ambiguous legal standard for developers and organizations. Unlike regulations that might set thresholds for computing power or model size, this effects-based definition places the burden on creators to prove their systems could not, under any circumstances, lead to humanity's disempowerment. This ambiguity is a significant departure from typical technology regulation. It raises critical questions about enforcement and interpretation: How would a prosecutor prove a system has such a capability before it is ever deployed? What level of evidence would be required to demonstrate that a model could not be misused for such ends? For engineering and legal teams, this would create a nearly impossible compliance environment, where the line between cutting-edge research and a felony is dangerously unclear.

This proposed ban does not exist in a vacuum but stands as a radical outlier in the global discussion on AI governance. Most regulatory frameworks, such as the European Union's AI Act, have adopted a risk-based tiered approach. These models categorize AI systems based on their potential for harm, imposing stricter rules for high-risk applications like medical devices or critical infrastructure, while allowing for more permissive development of low-risk tools. The Sanders-Casar bill rejects this nuanced approach in favor of a complete moratorium on a specific, albeit vaguely defined, category of AI. This hardline stance reflects the growing influence of the existential risk, or "x-risk," perspective within the AI safety community, which argues that the potential negative outcomes of superintelligence are so severe that its development should be halted entirely. The bill effectively brings this once-niche academic and philosophical debate directly into the legislative arena, forcing a national conversation on whether certain technological paths should be permanently closed off.

For founders, CTOs, and developers, the immediate practical impact of this bill is less about compliance and more about what it signals for the future of AI regulation. While the legislation faces a difficult path to becoming law in its current form, it establishes a new, more extreme anchor point in the policy debate. The introduction of potential prison sentences for AI development will likely force leading AI labs and their investors to take political and legal risk far more seriously. It could also have a chilling effect on open research and talent acquisition, as individuals may become hesitant to work on frontier models that could fall into a legal gray area. Moving forward, technology leaders must closely monitor the legislative climate, not just for practical rules on transparency and bias, but for fundamental questions about what kinds of AI are permissible to build at all. The key thing to watch is whether the concept of outright bans and criminal penalties gains traction and begins to influence more mainstream, bipartisan regulatory efforts.

This legislation marks a major shift from industry self-regulation to severe legal consequences for AI development. For developers and CTOs, it introduces the unprecedented risk of criminal charges for building advanced AI, potentially chilling research and setting a global regulatory precedent.

The proposed ban creates significant legal and compliance uncertainty for AI companies, especially those at the frontier of model development. Businesses may need to re-evaluate research roadmaps, increase legal budgets, and face new challenges in attracting talent for high-risk, advanced AI projects.