# On the AI Regulatory Commission > Source: > Published: 2026-09-21 11:34:31+00:00 1 I have a short essay today in the NY Times on the idea of an AI Regulatory Commission, which you can find here: https://www.nytimes.com/2026/09/21/opinion/bernie-sanders-jack-dorsey-ai-policy.html#link-5feefa9a This is not a small topic, of course. There are a ton of design details. Many people, including Tyler Cowen, have rightly drawn attention to the risks of unilateral action by the United States, and the importance of not falling behind China. (I address that question, sincerely if inadequately, in a prior Substack.) In case it might be of interest, here’s the ORIGINAL version of the essay I did for the NY TImes, produced with permission; the final version is significantly shorter and went through many edits. 2 In the coming months, Congress should create an Artificial Intelligence Regulatory Commission (AIRC). It should draw on existing models, including the Nuclear Regulatory Commission, the Environmental Protection Agency, and the Consumer Product Safety Commission. The AIRC could be a useful but relatively weak agency whose main duties involve investigation and factfinding. Alternatively, it could be an immensely powerful agency with the authority to make or break AI companies. The ideal approach would fall between the poles. Under that approach, the AIRC would consist of five commissioners, with bipartisan membership. Its first duty would be to report to Congress on existing risks – with, say, an initial report due within 120 days or its creation, and biannual reports thereafter. To produce its reports, it would be given the authority to investigate AI companies to examine current practices and to explore emerging dangers. The AIRC would also be authorized to issue emergency orders in the case of imminent risks. It would also be authorized to gather information from and to coordinate with agencies and leaders in other nations, in consultation with the Secretary of State and Secretary of Defense. It might be given rulemaking authority, to be exercised only after consideration of costs and benefits, and with a focus on reduction of catastrophic risks. It might be directed to impose fines for violations of the law. It might be authorized to approve of cooperative action, designed to promote prudence and care, that might otherwise raise antitrust issues. The AIRC would not be a sprawling bureaucracy; its staff should be streamlined and relatively small. In all cases (and this is essential), the AIRC would be subject to the legal constraints that generally face agencies and commissions of all kinds. These constraints would include requirements of public participation in rulemaking, including notice of proposed rules and an opportunity to comment, along with judicial review, designed to ensure its own compliance with the law and to police arbitrary action. New government agencies can, of course, do more harm than good, and there should be a presumption against them. There is a risk of agency capture; there is a risk of overreaching. The AIRC should not be authorized to squelch beneficial innovation -- or to allow other nations to move ahead of the United States. For those who are skeptical about the whole idea, it would make sense to consider a sunset provision, terminating the AIRC after two years unless Congress sees fit to reauthorize it. Still, AI poses an array of serious, poorly understood, and potentially growing risks: cyberattacks, theft, invasion of privacy, disruption of critical infrastructure (including financial services and energy supply), and uses of biological or other weapons, alongside even more catastrophic scenarios. The mere existence of an AIRC would put companies on notice and thus provide a measure of protection against those risks. And its limited but important authorities, as sketched here, could be deployed when necessary. What are we waiting for?