Nuclear-Level Penalties For Building AGI

Bernie Sanders speaking at a podium outdoors — Photo: lev radin / Shutterstock

Sen. Bernie Sanders moved to make building “superintelligent” AI a federal crime, with prison time on par with illicit nuclear work, signaling Washington may lock the brakes on the tech that could run the future.

Story Snapshot

Sanders introduced a bill to permanently ban superintelligent AI and pause advanced AI until safety rules exist.
The plan would create a new federal AI department and seek global deals to block superintelligence.
Violations could bring up to 20 years in prison, matching penalties for rogue nuclear development.
Supporters cite existential risk; some tech voices warn a broad pause could backfire.

What the Bill Would Do, in Plain Terms

Sen. Bernie Sanders and Rep. Greg Casar unveiled the Ban Artificial Superintelligence Act. The bill would permanently ban any AI that matches or beats human thinking across many fields. It would pause advanced AI work until a federal regulator sets safety rules. It would also push for global agreements to stop superintelligence abroad. The policy adds a new cabinet-level agency to police high-end AI. The press release and coverage outline these parts clearly.

The proposal defines target systems by broad capability, not by brand or model. That matters for enforcement. If a system can resist shutdown, run beyond human control, or outthink people widely, it would be illegal to build or deploy. The draft pairs the ban with criminal penalties for those who cross the line. Reports say sentences could reach 20 years in prison, which mirrors punishments for illegal nuclear weapons development under federal law.

Why This Fight Hits Nerves Across America

Sanders argues the risk is too great to leave to private labs and investors. He wants the United States government in the driver’s seat before an AI arms race cements winners and losers. Backers say the goal is to prevent systems that could escape human control. They also want public consent before any step toward superintelligence. Supporters link this to other high-risk tech where rules come first, including calls by leading scientists for a global curb on superintelligence.

Many readers, right and left, see a deeper issue. They fear powerful actors will shape AI to serve elites, not citizens. They worry about lost jobs, more surveillance, and another boom that lifts a few and leaves most behind. A centralized regulator could check corporate power. But it could also expand federal reach and invite capture by the same firms it oversees. The bill forces a core question: who decides how far AI goes, and on whose terms.

The Pushback: Pace the Frontier, Do Not Freeze It

Some industry leaders back slowing “frontier” AI but stop short of a total ban. OpenAI’s chief executive officer, Sam Altman, wrote that the frontier should be paced more slowly, aligning with calls to dial down speed rather than outlaw the end state. Others argue a pause now could raise risk by handing advantage to less transparent actors or hostile states. They ask what a pause would achieve without clear, testable milestones and global buy-in.

Reporters found Silicon Valley investors and builders split. Some accept strict guardrails for systems that could harm people at scale. Others doubt insider doomsday warnings and warn that a sweeping federal pause would slow useful tools, from medicine to education. Critics also flag hard lines to draw: what counts as “advanced,” who measures “across many fields,” and how to verify capability claims without enabling misuse during testing.

Enforcement, Global Gaps, and What Happens Next

Enforcing a capability-based ban will test Washington’s muscle. The new department would need to audit models, probe data centers, and track compute supply chains. It would need rules for red-teaming, export controls, and shutdown orders. It would also need penalties that bite but do not crush benign research. The bill’s prison terms show Congress is ready to use criminal law if civil fines cannot keep pace with the stakes in this field.

Bernie Sanders and Greg Casar unveiled the Ban Artificial Superintelligence Act today, permanently banning superintelligent AI and pausing advanced AI development until federal safety rules exist.

— Resistbot (@resistbot) September 23, 2026

Global gaps will matter as much as U.S. law. The bill calls for international deals to block superintelligence, but rivals may not sign on. If leading countries diverge, companies could relocate work, and open research could leak. That risk is why some experts press for a ban until there is broad scientific consensus on safe control and strong public support. Congress now faces a tight choice between precaution and innovation at a moment of rapid change.