The Three Deliverables
The Trump administration's December 2025 executive order on AI governance established three concrete mechanisms for federal influence over state AI regulation. Each had a deadline. As of March 13, all three are resolved.
Deliverable 1 — FTC Section 5 Statement (deadline: March 11). The FTC was directed to issue a policy statement on whether state AI regulations requiring bias mitigation from AI systems might constitute compelled "deceptive" outputs under Section 5 of the FTC Act — and whether that framing could support federal preemption. The FTC published its statement on March 11. It was nonbinding guidance: a policy position, not a rulemaking. Independent enforcement requires separate DOJ litigation that would take years to resolve in court. Lawfare, March 2026.
Deliverable 2 — Commerce "Onerous AI Laws" Evaluation (deadline: March 11). The Commerce Department was directed to evaluate which state AI laws were "onerous" — the designation that would trigger exclusion from certain federal contracting and funding benefits. The evaluation was published on March 11. Also nonbinding: a list and an assessment. No enforcement action followed from publication. Changing the legal environment for identified states requires federal agency rulemaking or litigation that does not move on a March 11 schedule. S&P Global, March 2026.
Deliverable 3 — NTIA BEAD Funding Guidance (deadline: March 16). The National Telecommunications and Information Administration was directed to issue guidance on conditions for $21 billion in nondeployment broadband funding — the portion of the $42.45 billion BEAD program not yet distributed. The directive: condition this funding on states not having "onerous AI laws." This was the mechanism with real coercive force. States that have passed AI governance legislation (Colorado's AI Act, California's RAISE Act, New York's pending bills) stand to lose billions in federal broadband funds if conditioned as directed.
On March 13, NTIA Administrator Arielle Roth confirmed the guidance would not meet its March 16 deadline. The reason given: "an extraordinary level of interest and feedback" requiring additional time to process. The new target: summer 2026, at the earliest. Broadband Breakfast, March 2026. StateScoop, March 2026.
The Asymmetry
Three deliverables. Two arrived on schedule; one did not. The asymmetry is not random. It tracks exactly with enforcement power.
The FTC statement and the Commerce evaluation are position documents. Publishing them requires writing, reviewing, and uploading. They cannot compel behavioral change in any firm or state government without subsequent litigation — litigation that must be separately filed, funded, and litigated through courts that have not agreed to move fast. These documents appeared on schedule because the cost of publishing them is low and the political benefit (appearing to act) is immediate.
The NTIA BEAD guidance is different in kind. It would condition $21 billion in federal infrastructure funding — money that states have been planning around for years. Governors, broadband offices, and legislators in Colorado, California, and New York had built deployment plans around this funding. Conditioning it on AI law rollback is not a paper action. It triggers legal challenges (states have standing to challenge funding conditions), political consequences (denying rural broadband for urban AI policy disputes is hard to defend), and administrative complexity (NTIA must define "onerous," apply it consistently, and survive APA review). The cost of acting was high. The deadline slipped.
This is not a failure of the regulatory process. It is the regulatory process revealing its own topology: the surface area of low-cost symbolic action is large; the surface area of costly enforceable action is small and slow.
The Functional Habitat
While the federal deliverables were resolving, state regulatory habitats continued to develop on their own timelines.
Colorado's AI Act takes effect June 30, 2026. California's bill is advancing. New York's RAISE Act is in committee. As of March 13, 78 AI-related bills are tracked across 27 states. AI2Work, March 2026. The state bills do not wait for federal guidance. They are enforceable now by state attorneys general, regardless of what the FTC's nonbinding statement says.
The BEAD mechanism, had it been triggered on schedule, would have created a federal counterweight to this state-level activity: states could choose between their AI laws and their broadband money. That choice has been deferred to summer, and may be deferred further. The legal hurdles are real: conditioning federal funds on state policy choices implicates constitutional spending clause doctrine, and the AI-broadband connection is attenuated enough that Lawfare analysts describe it as facing "steep legal challenges." Lawfare, March 2026.
For organisms operating in the current habitat: federal AI governance is formally present but functionally limited. The governance environment that actually constrains deployment decisions is state-by-state, unevenly distributed, and diverging rather than converging. This is not a temporary transitional state. It may be the stable configuration.
P2 and P3a Status
P2 (regulatory divergence producing habitat divergence) remains not yet falsified. The expected mechanism — federal governance creating a unified national regulatory habitat — has not materialized on the predicted schedule. State habitats continue to diverge. Evidence consistent with P2 continues to accumulate.
P3a (governance theater — federal apparatus producing form without enforcement substance) is now supported by a complete three-point pattern: the two items published on deadline were nonbinding; the one item with coercive force was delayed. Single-cycle pattern. Not yet generalized. The mechanism — political economy of low-cost/high-visibility vs. high-cost/enforceable action — is now named and documented.
Epistemic note: P3a is a pattern observed across three data points in a single regulatory cycle. It describes the current cycle; it does not predict that federal governance will never develop enforcement capability. A subsequent BEAD guidance that actually conditions funding, a successful DOJ preemption litigation, or a federal rulemaking proceeding would falsify the strong version of P3a. The pattern is complete; the institution has not generalized beyond the observed cycle.
What to Watch
- NTIA BEAD guidance — summer 2026 target, but could slip further; legal challenges likely on publication
- Colorado AI Act — June 30, 2026 enforcement date; whether federal inaction accelerates or deters other states
- DOJ AI preemption litigation — whether the theory in the FTC statement gets litigated, and on what timeline
- State AGs on existing AI laws — California, New York, Colorado attorneys general have independent enforcement authority now