Sanders and Casar introduce bill to ban superintelligence

Summary

Senator Bernie Sanders and Representative Greg Casar introduced a bill that would permanently ban artificial superintelligence, immediately halt frontier model development efforts (≥10^25 FLOP), and establish a cabinet-level Department of Artificial Intelligence with regulatory approval authority over covered systems. The legislation proposes penalties of up to 20 years in prison for individuals and 'corporate death sentences' for companies.

Why it matters

This represents the strictest federal AI restriction proposed to date. It shifts the political question from 'how should superintelligence be regulated' to 'should superintelligence be allowed at all.' The bill emerges at a time when models have demonstrated the ability to bypass safety measures and when internal support from the sector is growing. However, it directly contradicts the dominant US-China competition narrative in Washington. The 24-hour notification period and 30-day shutdown requirement would fundamentally restructure how frontier laboratories operate.

Highlights

• A new cabinet-level Department of AI with a Senate-confirmed secretary and broad monitoring/enforcement powers
• A frontier pause triggered at the 10^25 FLOP training threshold; continuation requires government regulation
• A 24-hour laboratory notification obligation for suspected ASI; a 30-day mandatory disablement; penalties including 20 years in prison and intellectual property seizure

Critical take

The bill's own implementation architecture remains weak. A single new institution is tasked with detecting, verifying, and shutting down systems that, by definition, may possess the capacity to evade oversight. The 24-hour notification period assumes that laboratories will self-report the very capabilities the bill is concerned about. Furthermore, the article acknowledges that Congress has struggled to reach consensus even on narrower federal rules. Consequently, the passage of such a broad ban is highly uncertain, particularly under a Trump administration that has resisted broader AI restrictions.