Bernie Sanders Wants 20 Years In Prison For Building AI That Goes Too Far

Bernie Sanders and Greg Casar have introduced one of the most aggressive AI proposals yet seen in Washington, and it comes with a penalty that immediately grabs attention: up to 20 years in prison for people who violate the proposed restrictions. Their new bill would permanently ban the development or deployment of artificial superintelligence while placing a temporary pause on certain advanced AI development until a new federal regulatory system is established.

The legislation would also create a Cabinet-level Department of Artificial Intelligence with authority to monitor advanced systems, oversee the removal of dangerous capabilities and supervise the destruction of prohibited superintelligence. Sanders and Casar argue that increasingly powerful AI is developing faster than government oversight can keep up, while critics say the proposal is too broad, difficult to define and unlikely to survive the political and legal battles ahead.

Sanders And Casar Are Calling For A Major AI Freeze

The Ban Artificial Superintelligence Act is aimed at a very specific concern: the possibility that AI systems could eventually become capable of operating beyond meaningful human control. The proposal would prohibit the development and deployment of artificial superintelligence while temporarily pausing advanced AI development until federal safety rules and a model review system are established.

The lawmakers introduced the legislation on September 23 after weeks of warnings about the risks associated with increasingly capable AI. Sanders has argued that government needs to intervene before systems become too powerful to control, rather than waiting for an irreversible incident to force action. The proposal therefore treats advanced AI development as an issue requiring direct federal oversight rather than relying primarily on voluntary commitments from technology companies.

Sanders described the technology as moving at extraordinary speed and argued that decisions about its future should not be left entirely to a small group of wealthy technology executives. The bill reflects that position by placing a new federal agency in charge of setting standards, reviewing advanced systems and enforcing restrictions on capabilities considered dangerous.

The 20-Year Prison Penalty Is The Part Getting Attention

Under the proposal, companies that violate or attempt to circumvent the restrictions could face what the lawmakers describe as a “corporate death penalty.” Individuals could face a maximum sentence of 20 years in prison. Sanders’ office says that proposed prison term is similar to existing penalties associated with unlawfully developing nuclear weapons.

The punishment would apply under the proposed legislation, meaning it is not currently a federal criminal penalty for developing advanced AI. The bill would first have to pass Congress and become law before its provisions could be enforced. That distinction is important because the proposal is currently a legislative effort rather than an enacted ban on AI development.

The severity of the proposed punishment reflects how seriously Sanders and Casar view the potential risks. Their proposal does not simply call for companies to disclose how their models work or conduct additional safety testing. It would establish criminal consequences for certain prohibited development and give the federal government much broader authority over the most advanced systems.

What Would Actually Count As Superintelligence?

Defining the technology may prove harder than announcing a ban. The proposal describes artificial superintelligence in terms of systems that exceed human cognitive performance across most domains or possess sufficient capabilities to destroy or disempower humanity. The bill also identifies dangerous capabilities that could include defeating shutdown commands, conducting unauthorized cyberattacks or overthrowing the federal government.

One particularly significant concern is the ability of AI to develop or improve AI systems with limited human involvement. Sanders and Casar’s proposal specifically targets capabilities that could automate or greatly accelerate AI research and development. The concern is that an AI system capable of rapidly improving the technology behind itself could make the pace of development much harder for humans to control.

That creates a difficult regulatory problem. AI capabilities exist along a spectrum, and determining when a system crosses from powerful software into legally prohibited superintelligence would require technical standards that can withstand constant changes in the technology. The proposal attempts to focus on what an AI system can actually do, but the boundaries could become a major point of debate if lawmakers consider the legislation.

A New Department Of Artificial Intelligence Would Police It

The bill would establish a new Cabinet-level Department of Artificial Intelligence to oversee the technology. The department would be responsible for creating safety rules and model review processes before advanced AI development could resume under the proposed framework. It would also monitor frontier AI systems throughout their lifecycles for dangerous capabilities.

The proposed agency would have unusually broad responsibilities. Its scientists and officials would monitor advanced systems, supervise the removal of dangerous capabilities and oversee the destruction of artificial superintelligence that falls under the ban. An Artificial Intelligence Advisory Board would also provide scientific and technical advice to the department.

The legislation would push the federal government into a much more direct role in deciding which AI systems can be developed and deployed. It would also seek international agreements and allied coordination designed to prevent prohibited superintelligence from being developed elsewhere, including through policies such as export controls.

Supporters Say Waiting Could Be Riskier

Sanders and Casar have built their case around the possibility that AI development could eventually move beyond effective human oversight. In September, Sanders argued that increasingly powerful systems should be paused before they reach a point where people can no longer reliably control them. Casar has similarly warned that advanced AI could create serious risks to security, freedom and human life.

The lawmakers’ argument has gained support from some people in the AI safety community who favor stronger government intervention. Their position is that safety testing and independent oversight should happen before systems become capable of causing irreversible harm, rather than after a serious incident has already occurred.

That debate has become more urgent as researchers and technology companies work toward systems capable of performing increasingly sophisticated tasks. Recent developments have included AI models being used for autonomous research, cybersecurity and other activities that previously required significant human involvement. The central disagreement is over how quickly those capabilities could advance and how much risk regulators should accept while development continues.

Critics Say The Proposal Could Be Too Broad

The legislation has also drawn criticism from people who believe its definitions and penalties are too sweeping. One industry representative described the bill as effectively amounting to an AI ban, arguing that some companies could already meet parts of the proposed threshold depending on how the language is interpreted.

Critics have also questioned whether the proposed penalties would withstand constitutional challenges. Those concerns have not been resolved because the bill has only been introduced and has not gone through the legislative process, much less been tested in court.

The larger disagreement is about what kind of regulation should govern advanced AI. Some lawmakers want mandatory safety testing and independent verification, while others favor lighter regulation that gives companies more freedom to develop the technology. Sanders and Casar have chosen a far more restrictive approach by proposing an outright prohibition on artificial superintelligence.

Congress Has Several AI Plans On The Table

The Sanders-Casar bill is arriving alongside several other attempts to address AI risks. Sen. Maria Cantwell has called for federal standards, independent testing and international coordination, including mechanisms that could allow governments to communicate about severe AI incidents. Those proposals would regulate the technology without necessarily imposing the permanent ban sought by Sanders and Casar.

Sen. Ted Cruz is also working with other senators on legislation addressing catastrophic AI risks, including potential biological and nuclear threats. Another proposal would require AI companies to maintain an emergency “kill switch” that could shut down systems during a serious incident. That effort previously failed to advance after Sen. Rand Paul objected to moving it forward by unanimous consent.

The competing approaches show how unsettled the federal AI debate remains. Lawmakers broadly face the same basic question, but the proposed answers range from emergency shutdown requirements and independent testing to sweeping restrictions on development and a permanent ban on artificial superintelligence.

The Bill Still Faces A Long Road

For all the attention surrounding the proposal, the 20-year prison penalty is not about to become law. Sanders and Casar have introduced legislation, and it now faces the normal congressional process before any of its restrictions could take effect. The proposal would need enough support in both chambers before reaching the president.

The timing also creates another obstacle. Congress is approaching the 2026 midterm elections, leaving lawmakers with limited time to move major legislation. Other senators are still negotiating their own AI proposals, and there is no indication that Congress has reached a consensus on a single regulatory framework.

For now, the Sanders-Casar bill has done something that few AI proposals have managed to do this dramatically: it has put the possibility of criminal punishment for developing certain AI systems directly into the congressional debate. The technology may keep advancing, but Washington is now being forced to decide where the legal line should be drawn before something crosses it.