US: Sanders and Casar propose licences for developing the most powerful AI models and a ban on superintelligence
Independent Senator Bernie Sanders and Democratic Congressman Greg Casar introduced the Ban Artificial Superintelligence Act (S. 5493 and H.R. 10538) on 23 and 24 September 2026. The bill would establish a federal Department of Artificial Intelligence. Advanced AI systems could be developed or distributed only by those holding a charter issued by the Secretary. An advanced system is one trained on computing power of at least 10²⁵ operations; the Secretary is to adjust this threshold annually to reflect changes in training efficiency. A charter holder must give the Secretary access to its systems, processes, staff and physical infrastructure.
Until the Department has issued its rules, a mandatory pause would apply: systems above the threshold could not be further trained or fine-tuned, and unreleased ones could not be deployed. Every release would then require prior approval. The bill bans the development of artificial superintelligence outright. Reckless violations would expose individuals in policymaking roles to up to 20 years in prison and other employees to a ten-year ban from working in the AI industry. Companies would face revocation of their charter and forfeiture of their intellectual property and assets.
Article 51(2) of the AI Act uses the same threshold of 10²⁵ FLOP. According to the Commission, such models are currently developed by only a handful of companies and training one costs tens of millions of euros, so the licensing requirement would today apply in practice only to the most powerful models of a few companies. In the EU, however, the threshold only triggers a presumption of systemic risk, with a notification duty and the obligations under Article 55, not a licensing requirement.
The Senate bill was introduced by Sanders alone; the House bill has 12 Democratic cosponsors, including Alexandria Ocasio-Cortez. Both bills have so far only been referred to committee. If Congress does not pass them before the end of its term on 3 January 2027, they will lapse.
Affected provision
More news
- The implementing regulation on AI regulatory sandboxes has been published: free of charge for SMEs and start-ups, with exit reports recognised across the EU
- EDPB: the supervisory authorities' coordinated enforcement action in 2027 will target automated decision-making and AI in recruitment
- The Czech Government has approved the AI Act implementing bill and set up a Government Council for Artificial Intelligence