A pair of lawmakers introduced sweeping restrictions on artificial intelligence development this week, proposing to criminalize the creation of superintelligent systems while imposing a temporary halt on cutting-edge AI research until federal safety standards take shape.
Sen. Bernie Sanders of Vermont and Rep. Greg Casar of Texas jointly unveiled the Ban Artificial Superintelligence Act, marking one of the most aggressive regulatory approaches to AI safety yet proposed in Congress. The bill targets what researchers call superintelligence: AI systems that would surpass human-level capabilities across most cognitive domains.
Core Provisions
The legislation establishes three main components, according to AI Weekly. First, it would permanently prohibit the development and deployment of superintelligent AI systems. Second, it would create a temporary moratorium on advanced AI research until a newly formed federal regulator develops comprehensive safety standards. Third, it would establish a cabinet-level agency responsible for overseeing artificial intelligence across all sectors of the economy.
Enforcement and Penalties
The bill includes substantial criminal penalties designed to deter violations. Individuals found developing prohibited superintelligent systems would face up to 20 years in prison. Corporate entities would encounter what the lawmakers characterize as capital penalties, effectively allowing for corporate dissolution as punishment for non-compliance.
This enforcement structure represents a significant escalation from existing regulatory frameworks, which typically rely on fines or operational restrictions rather than incarceration or corporate dissolution.
The Safety Debate
The proposal reflects growing concerns among policymakers about the trajectory of AI development. Proponents argue that superintelligent systems pose existential risks that demand preemptive legal barriers. The bill essentially treats superintelligence as a category of technology too dangerous to permit, similar to how certain biological weapons or nuclear materials are regulated.
Critics of such restrictions counter that the definition of superintelligence remains unclear and that overly restrictive approaches could chill beneficial research while proving difficult to enforce internationally. They note that AI capabilities are distributed globally across multiple jurisdictions, making unilateral bans potentially ineffective.
Regulatory Architecture
The cabinet-level oversight agency proposed in the bill would hold authority to establish safety standards before permitting continued development of frontier AI systems. This approach differs from existing regulatory bodies like the Federal Trade Commission, which enforce rules retroactively after potential harms occur.
The pause on advanced AI research would remain in effect until this new regulator certifies that safety protocols are adequate. Such a mechanism would effectively give federal authorities power to slow or accelerate AI development based on their assessment of risk.
What's Next
The bill faces an uncertain path through Congress. Tech industry groups and research institutions have traditionally opposed broad development bans, arguing they could disadvantage American companies relative to international competitors. Supporters of the legislation, however, argue that safety concerns outweigh competitive considerations.
The introduction of this bill signals that some lawmakers view voluntary industry safety commitments as insufficient and believe statutory prohibition may be necessary to manage risks associated with advanced AI systems.
