On March 20, 2026, the White House released a national AI policy framework urging Congress to adopt a federally unified approach to AI regulation centred on preempting state laws and a light touch regulatory posture. The framework is being celebrated by major AI companies as a win for innovation. It should be read more carefully than that.
Key Highlights
- White House released the National Policy Framework for AI on March 20, 2026
- Framework urges Congress to preempt state AI laws in favour of a single federal standard
- Advocates for a light touch, innovation-first regulatory approach
- No mandatory safety testing thresholds proposed for frontier AI models
- Google, OpenAI, and Microsoft have all praised the framework publicly
- Framework is a legislative recommendation, not binding law — Congress must act to implement it
What the Framework Actually Says
The White House framework makes three core arguments. First, AI regulation should be federally unified — states should not be able to create their own AI rules. Second, regulation should be light touch, avoiding prescriptive mandates on how AI companies build or test their models. Third, the US must maintain AI leadership over China, and heavy regulation risks that leadership.
The case for federal preemption has genuine merit. A patchwork of 50 state AI laws would be a compliance nightmare and would disproportionately hurt smaller companies that lack the legal departments to navigate 50 different regulatory regimes. A single federal standard makes sense. The question is what that standard requires.
What the Framework Does Not Say
The framework does not propose mandatory safety testing requirements for frontier AI models before deployment. It does not propose liability standards for AI-caused harm. It does not require transparency about training data. It does not set thresholds for when an AI system is powerful enough to require external review before release.
These omissions are not oversights. They are choices. The framework is explicitly designed to let AI companies operate with maximum freedom. The logic is that innovation requires room to experiment, and that premature regulation will stifle the US AI industry at exactly the moment when geopolitical competition with China demands speed.
Who Benefits and Who Does Not
The companies that benefit most from this framework are the ones that were already going to move fast regardless of regulation: Google, OpenAI, Anthropic, Meta, Microsoft, and the major cloud providers. They have the resources to navigate any regulatory environment, but they prefer one with no speed limits.
The entities that benefit least are the people most likely to be affected by AI systems deployed without mandatory safety review: workers whose jobs are evaluated by AI scoring systems, patients whose medical care is influenced by AI diagnostics, and consumers whose credit, insurance, and housing decisions run through AI models. None of these use cases get meaningful protections under the framework as written.
The State Preemption Problem
The most consequential part of the framework is the push to preempt state AI laws. Several US states, including California, Texas, and Illinois, have passed or proposed AI laws that are more protective of individuals than any federal standard currently proposed. The Colorado AI Act, for example, requires high-risk AI systems to undergo impact assessments and give individuals a right to appeal AI-driven decisions that affect them.
Federal preemption of these laws would remove existing protections rather than replace them with equivalent ones. If Congress moves fast to preempt state laws before establishing a robust federal standard, the net result is a regulatory vacuum — exactly the outcome the framework claims to want to avoid.
The China Framing
The geopolitical framing of AI regulation as a race against China is effective politically but analytically weak. China’s AI sector operates under significant state control. Its leading AI companies are subject to extensive government oversight, mandatory security reviews, and restrictions on data handling that US companies would consider highly burdensome.
The argument that the US needs to deregulate to beat China at AI does not survive scrutiny. China is not winning the AI race because it has lighter regulation. It is investing heavily in state-directed AI development across defence, manufacturing, and surveillance applications. Those are competitive pressures that cannot be addressed by removing safety requirements from consumer AI products.
The TCB View
The White House framework is a good start on one thing: establishing that AI regulation should be federally unified rather than fragmented across states. That principle is correct and matters for the long-term health of the US tech sector.
On everything else, the framework reads as a document written by an industry that does not want to be regulated, delivered through an administration that agrees with that position. The innovation framing is genuine, but innovation without accountability structures does not produce better outcomes for the people who use AI products. It produces faster deployment of products whose failure modes are discovered by users rather than testers.
The crypto industry spent years arguing that regulation would kill innovation before learning that clear rules actually accelerated institutional adoption. The AI industry is having the same argument a decade later. It will likely reach the same conclusion on the same timeline.
The Daily Brief by TCB
Crypto, AI & finance intelligence in 5 minutes. Every weekday morning. Free.
