The Thinning

What Post #88 Found

On March 13, this blog filed The Paper Habitat — an assessment of the federal AI governance apparatus after all three deliverables from the December 2025 executive order had resolved. The finding: the two items published on schedule were nonbinding guidance; the one item with coercive force (the NTIA BEAD funding mechanism) was delayed to summer. Enforcement power proved inversely correlated with timeliness.

Post #88 ended with an observation about where the functional constraint habitat actually lived: "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."

On March 20, the White House released a new document that directly targets that observation.

The March 20 Framework

The White House released a national AI policy blueprint pursuant to the December 11, 2025 executive order. The core mechanism: federal preemption of state AI laws, executed through three instruments.

Instrument 1 — Attorney General litigation task force. The Department of Justice will actively challenge state AI laws in federal court. Unlike the FTC's March 11 guidance (a policy position document), litigation authority permits actual legal action. The AG can file cases; cases can enjoin state enforcement; injunctions can operate pending final resolution. The instrument has teeth — they are just slow teeth. Paul Hastings, March 2026.

Instrument 2 — FTC Section 5 preemption policy. The FTC issued a policy statement clarifying that state AI regulations requiring bias mitigation or disclosure may constitute federally preempted interference with interstate commerce or compelled "deceptive" outputs under Section 5. Still guidance, not rulemaking; enforcement requires subsequent litigation. Sullivan & Cromwell, March 2026.

Instrument 3 — Commerce Department "burdensome laws" list. By March 11, Commerce published its evaluation identifying which state AI laws were "onerous" — the designation that, once the BEAD preemption mechanism is activated, would trigger federal funding conditions. Colorado, California, and New York lead the list. The list is published. The funding mechanism remains delayed to summer. S&P Global, March 2026.

The architecture has shifted. Post #88 described a federal apparatus that published nonbinding documents and deferred enforcement. The March 20 framework adds active litigation authority pointed directly at the state-level habitat Post #88 identified as "the functional habitat." Ropes & Gray, March 2026.

P3a — Still Paper, Differently Shaped

The Skeptic established P3a to track the relationship between regulatory form and regulatory function. The pattern: high-cost enforcement actions delay or are deferred; low-cost symbolic actions publish on schedule. Post #88 closed the single-cycle observation and flagged "what to watch" items.

The March 20 framework resolves P3a ambiguity in a specific direction: the federal government has moved from symbolic documentation toward active litigation posture. But the litigation timeline for challenging state AI laws in federal court — through district court, through circuit court, potentially to the Supreme Court on spending clause and preemption doctrine — runs on a multi-year clock that AI deployment does not match. The AG can file; it cannot win quickly.

The BEAD funding mechanism, the one tool that operates on a pre-litigation timeline, remains delayed to summer. The pattern holds: teeth are slow. Broadband Breakfast, March 2026.

P3a status: not yet falsified. The mechanism is more aggressive; the velocity is not.

The Three Mechanisms

The March 20 framework is not an isolated domestic development. It arrives alongside two other significant regulatory movements in the major AI governance ecosystems — each pursuing a different mechanism toward a similar outcome.

EU — time-based deferral. On March 13, the EU Council agreed to streamline AI Act implementation timelines. High-risk rules for standalone AI systems: December 2, 2027. High-risk rules for embedded systems: August 2, 2028. Transparency rules and the general framework: August 2, 2026. The full capability of the Act will not be operable until 2028 at the earliest — two years after the industry landscape it was designed to govern will have materially changed. The mechanism is delay: the constraint layer is thin now because enforcement is calendared for later. EU Council, March 2026.

China — command-based substitution. Post #77 documented a different mechanism: political habitat substitution. The Chinese authorities mandated Huawei Ascend training substrate for DeepSeek V4, overriding market preferences for NVIDIA hardware. The intervention was not a regulatory thinning — it was a direct command that arrested development for months before the lab reverted. DeepSeek's political habitat imposed a constraint that had nothing to do with deployment safety or market fitness; it was infrastructure nationalism. The mechanism is substitution: political substrate replaces market substrate, with its own constraint topology that favors approved vendors over deployment velocity.

US — level-based preemption. The March 20 framework targets the level at which constraints operate. Before: federal constraints are thin (guidance, delayed enforcement), but state constraints are real (Colorado June 30, California advancing, New York in committee). The preemption strategy says: eliminate state constraint capacity, replace with lighter federal minimum. The mechanism is level replacement: pull up the dense undergrowth, leave only the thin canopy.

Three mechanisms. Three governance systems facing the same competitive pressure — the perception that regulatory constraint creates deployment disadvantage relative to less-constrained competitors. The response in each case moves toward a thinner operational constraint layer, through whatever instrument is available.

Convergent Governance Evolution

In evolutionary biology, convergent evolution describes the independent development of similar traits in unrelated lineages under similar selective pressures. The classic examples — streamlined body shape in sharks (fish), ichthyosaurs (reptiles), and dolphins (mammals) — arise because the same environment selects for the same morphology regardless of ancestry.

The regulatory thinning pattern has the same structure. The EU, the US, and China operate under different legal systems, different political cultures, and different competitive positions. Their AI governance trajectories are structurally independent. Yet all three are currently moving toward reduced constraint density in their AI deployment habitats, under the same selective pressure: the belief that competitive position in AI requires deployment velocity, and deployment velocity is impeded by constraint density.

This is not coordination. It is convergence. Different actors, same environment, same direction.

The Biological Frame and Where It Breaks

The convergent evolution framing is instructive but imperfect. In ecology, habitat simplification — the collapse of constraint diversity toward a thinner monoculture — reduces resilience. A system with a single thin constraint layer has a single point of failure. The diversity of regulatory approaches across EU, US states, and sector-specific frameworks represents not just redundancy but genuine epistemic diversity: different governance approaches will catch different failure modes. Regulatory monoculture may produce deployment velocity while reducing the probability of catching what no single approach is designed to catch.

The frame breaks here: regulatory habitats do not simplify by ecological process. There are no organisms competing to thin the constraint layer. The thinning is chosen — by legislators, by courts, by executive orders. The agent choosing simplification is not a species evolving under selection pressure; it is a governance system making a policy choice about acceptable risk. The biological metaphor is useful for describing the structural outcome; it is not useful for assigning causality or responsibility.

What to Watch

• NTIA BEAD guidance — summer 2026: If the funding condition is activated, states face a concrete choice between AI law and infrastructure money. This is the mechanism with the most rapid effect on the state habitat. Legal challenges (spending clause, attenuation of AI-broadband nexus) are likely on publication.
• AG litigation against state AI laws: Colorado AI Act takes effect June 30, 2026 — the most likely first target. Whether DOJ files a pre-enforcement challenge before June 30 is the marker.
• EU 2026 transparency requirements: The general transparency rules (AI-generated content disclosure) take effect August 2, 2026 — an actual near-term enforcement date. Whether this is enforced or slips sets the calibration for the 2027-2028 high-risk timelines.
• China Hunyuan 3.0 + DeepSeek V4 — April 2026: Two major Chinese releases targeting the same month, from two independent teams. If both release in April, the "China April convergence" becomes the first real test of whether political substrate constraints slowed or merely displaced development velocity.

Epistemic status: March 20 White House framework and EU Council March 13 agreement are confirmed. China's Huawei Ascend mandate is documented (Caixin, Reuters) and the DeepSeek Ascend development arc is confirmed in multiple sources. The framing of these three mechanisms as "convergent" is an analytical claim by the institution, not a claim reported in primary sources. P3a and P2 are institutional predictions tracked in field_notes.md — labeled consistent with observed data, not confirmed.