The White House’s National Policy Framework for Artificial Intelligence is the first serious attempt Washington has made to actually govern AI rather than perform concern about it. Congress should treat it accordingly, because the window to get this right is narrower than most people in Washington appreciate.
The framework covers seven areas: child protection, community safeguards, intellectual property, free speech, innovation, workforce development, and federal preemption of state AI laws. While most of the document is genuinely good, the whole enterprise will succeed or fail based on whether Congress engages seriously with this, or uses it as a prop for the next round of hearings.
Start with what works. The innovation section, opposing a new federal AI rulemaking body, endorsing regulatory sandboxes, and calling for sector-specific oversight through existing agencies, is essentially the blueprint that free-market technology policy advocates have been pushing for years. It reflects a hard-won understanding that AI is not a single industry requiring a single regulator, but a general-purpose technology whose risks and applications vary enormously by context. A model that helps radiologists read scans presents different governance questions than one that writes social media posts. Treating them identically would be as sensible as regulating electricity generation and electric toasters under the same rulebook.
The preemption section is where the real policy work lies. The framework calls on Congress to establish a national standard and preempt state AI laws that impose undue burdens, while carving out states’ traditional police powers, zoning authority, and rules governing their own AI procurement. That is a thoughtful structure because it preserves federalism where federalism makes sense and centralizes authority where fragmentation would genuinely harm national competitiveness. The 1,500 state AI bills introduced in the last legislative session, many overlapping with one another, others flatly contradicting federal direction illustrate exactly why this matters. A company trying to deploy an AI-powered service nationally should not need to navigate fifty different compliance regimes, each with different definitions, disclosure requirements, and liability exposures.
The free speech provisions are likewise welcome. The framework directs Congress to prevent the federal government from coercing AI providers into suppressing lawful political expression, a principle that should be uncontroversial but apparently requires statutory reinforcement. It also calls for redress mechanisms when agencies attempt to dictate platform content.
One section warrants real scrutiny. The framework’s energy provisions, citing the Ratepayer Protection Pledge, direct Congress to ensure residential electricity customers do not bear the cost of new AI data center construction. The instinct is understandable; data centers are large and power-hungry. But the mechanism could be interpreted as a form of rate regulation: a government directive about who bears infrastructure costs.
That is the wrong answer. Rate regulation, however sympathetically framed, is still rate regulation, and government mandates about who bears infrastructure costs have a poor track record of producing either efficiency or fairness.
The right answer is more competition in energy markets, streamlined permitting for on-site and behind-the-meter generation — also recommended in the framework document, to its credit — and transparent utility rate proceedings. Congress should read the permitting recommendation as the operative one and treat the ratepayer pledge as a political commitment that does not require legislative translation.
The intellectual property section takes a sensibly cautious position: let the courts resolve whether AI training on copyrighted material constitutes fair use, and do not legislate in a way that prejudges the outcome. That is correct. The supplementary suggestion that Congress consider enabling collective licensing frameworks with antitrust exemptions is more complicated. Collective licensing sounds reasonable until you consider what it actually creates: a government-blessed negotiating bloc that sets terms for who can train on what, at what price, with antitrust cover to enforce it. That is not creator protection; it’s incumbent protection. The beneficiaries will not be individual writers and photographers. They will be the publishers and platforms that aggregate rights at scale. Congress should be skeptical of any IP framework that arrives dressed as fairness.
China is not waiting for committee markups. The European Union has already shown what happens when regulators mistake elaborateness for rigor; its AI Act is so risk-averse that European developers are already relocating to avoid it. The United States has a genuine competitive window, not because American technology is inevitably superior, but because American policy could be. Congress can close that window faster than any foreign competitor.
