Bernie Sanders and Greg Casar introduced the Ban Artificial Superintelligence Act on September 23, 2026, according to their offices. They had announced the proposal earlier in September. It is not enacted law.
The bill would create a federal Department of Artificial Intelligence. Advanced AI development would pause until the department is staffed and has established safety rules. Future releases of those systems would require government approval.
It also proposes banning superintelligence and specified precursor capabilities. These include unauthorized access, evading shutdown and independently improving a system. They are proposed legal categories, not evidence that every current model has those capabilities.
The Associated Press confirms the proposal and describes difficult prospects in Congress. The introduction does not establish an international agreement or put a development pause into effect.
Organizations concerned about loss of control disagree on the details. ControlAI welcomes the ban's goal but considers the broad pause and treatment of precursor capabilities excessive. The Alliance for Secure AI credits the initiative while calling for a more measured approach.
We record this as a Risk Signal: a concrete oversight proposal and the debate around it. It is neither a new AI incident nor a measure of catastrophe probability.