By Hamza Faisal

Democrats in Congress are rallying behind one of the most aggressive AI regulation proposals ever drafted: a bill that would pause the training of advanced AI systems, force labs to get a federal licence before building them, and send violators to prison for up to 20 years.

The legislation, the Ban Artificial Superintelligence Act of 2026, has gained fresh momentum since the November 3 midterm elections handed Democrats control of the House. Arizona Representative Yassamin Ansari summed up the mood in a November 4 post, writing that a pause on advanced AI development, regulation of the industry, and a ban on artificial superintelligence should be "top of the Democratic agenda on day one".

The bill is being pushed by Representatives Greg Casar and Alexandria Ocasio-Cortez in the House, and Senator Bernie Sanders in the Senate. Sanders introduced the Senate version, S.5493, on September 23; Casar introduced the House version, H.R.10538, a day later with nine original co-sponsors.

What the bill would do

At the centre of the proposal is a new cabinet department: the Department of Artificial Intelligence, led by a secretary nominated by the president and confirmed by the Senate. Until that department is fully staffed and has written its safety rules, the bill would impose a mandatory pause on training, modifying or fine-tuning any AI system trained using 10^25 operations or more — a compute threshold designed to capture only the largest frontier models, adjustable over time as training methods become more efficient.

Once the department is running, no company could develop or distribute an advanced model without a federal charter. Labs would have to submit to inspections of their systems, processes, personnel and facilities, and win pre-deployment approval before releasing a covered model. The bill would permanently ban artificial superintelligence and systems built to develop it.

The penalties are unusually severe for technology legislation. Individuals in decision-making roles, and outsiders who help a company break the rules, could face up to 20 years in federal prison for reckless violations. Other workers at offending companies could be barred from the industry for a decade, and a company's charter could be revoked.

Why the push is happening now

The renewed push comes as the AI boom has run for years with almost no federal oversight. The incoming Democratic majority appears keen to change that: alongside the superintelligence bill, party lawmakers are weighing rules on data-centre electricity use and on liability for AI products that cause harm.

Supporters say frontier models are advancing faster than safety science, and a pause is the only way to give regulators time to catch up. Critics warn a training pause would hand the lead in AI to China and punish American labs for work that is already closely watched.

Long odds of becoming law

For now, the proposal remains just that — a proposal. Both versions sit in committee, with no hearing scheduled and no vote taken. To become law, the bill would need to pass the House and the Senate and be signed by the president, or survive a veto with two-thirds of both chambers. It also expires when the current Congress ends on January 3, 2027, unless it is reintroduced.

Senators Michael Bennet and Peter Welch have a separate, milder proposal: an independent Federal Digital Commission to oversee frontier models and major digital platforms instead of a cabinet department.

Whether more moderate Democrats will endorse the pause remains an open question. But if it ever passed, it would transform an industry that has so far operated without any higher authority in the United States.