Commentary. The political argument over artificial intelligence has entered a more serious phase. After the November 2026 US midterm elections, incoming lawmakers and state leaders are pressing for tougher oversight of AI systems and the data centres that power them. The shift reflects real public anxiety—but it should not become an excuse for symbolic bans that regulators cannot enforce.
Our view is straightforward: governments should regulate AI more aggressively, but they should regulate how systems are built and used rather than trying to halt the technology wholesale. The case for action is no longer theoretical. States are filling the gap left by the absence of a comprehensive federal framework, while companies deploy increasingly capable systems across healthcare, policing, education and finance. A patchwork of rules is better than no rules, but it is a poor substitute for clear national standards.
The political pressure is visible in the proposals now emerging. Illinois Gov. JB Pritzker has called for federal privacy legislation, mandatory AI risk assessments and requirements that data centres use energy and water efficiently. Other incoming lawmakers have discussed pausing development of advanced systems until a new federal authority establishes safety rules. Meanwhile, policy specialists report that the United States still lacks a broad national regulatory framework.
These proposals deserve more than industry dismissal. AI can amplify discrimination, expose private data, automate mistakes and concentrate power in a handful of companies. Data centres also consume substantial electricity and water, placing costs on communities that may receive relatively few permanent jobs. Requiring testing, incident reporting, independent audits and resource disclosures is not anti-technology. It is a basic condition for public trust.
Yet a blanket pause would be the wrong instrument. A moratorium imposed in one country could be difficult to police when development, deployment and investment cross borders. It could also favour the largest firms, which are best equipped to absorb compliance costs, while pushing smaller researchers and startups out of the market. Most importantly, a training threshold alone would miss dangerous applications built from smaller models or assembled into automated agents.
“The argument is not whether AI should be governed. It is whether governance will be practical enough to protect people outside the conference room.”
The better path is a risk-based framework. High-impact uses—medical decisions, employment screening, law enforcement, critical infrastructure and systems handling sensitive personal data—should face strict pre-deployment evaluation and continuing monitoring. Lower-risk tools should encounter lighter requirements, so that regulation does not turn every spreadsheet assistant into a licensing project.
Accountability must follow the system through its life cycle. Developers should document training data and known limitations; deployers should assess local harms; and both should preserve records that allow investigators to reconstruct what went wrong. Regulators need technical expertise, secure access to evidence and the authority to impose meaningful penalties. Voluntary promises can supplement those powers, but they cannot replace them.
Energy and water rules require similar balance. Communities are right to demand that data-centre projects disclose consumption, fund grid improvements and avoid shifting environmental burdens onto residents. But national policy should also prevent every planning dispute from becoming a de facto veto over digital infrastructure. Efficiency standards, transparent permitting and enforceable local benefit agreements offer a more durable compromise than either unrestricted construction or blanket bans.
International coordination will matter because neither algorithms nor corporations respect national borders. Recent policy discussions in Europe, Asia and the United States show growing interest in common obligations, but cooperation must produce compatible testing and reporting standards rather than another layer of aspirational declarations.
The public has good reason to distrust promises that safety will arrive later. At the same time, the public interest is not served by pretending that technological retreat is simple or cost-free. AI may deliver useful gains in medicine, science and productivity, and responsible rules should preserve those benefits while limiting concentrated risks.
That is the standard policymakers should now apply: regulate consequences, demand evidence and make powerful actors answerable. A freeze may sound decisive. A durable system of enforceable guardrails would be more useful.