On the evening of December 11, President Trump signed an executive order titled “Ensuring a National Policy Framework for Artificial Intelligence.” It declares a policy of a “minimally burdensome national policy framework for AI” and sets several parts of the federal government to work against state AI laws the administration considers onerous, until Congress passes a single national standard.
The order drew immediate reaction. Route Fifty quoted Rep. Zoe Lofgren saying the President “can work with Congress on both sides of the aisle to debate and pass a federal standard,” while industry groups including ITI and NetChoice welcomed the push for national rules.
/ 01What the order does
The order works through deadlines and federal levers rather than a single preemption clause:
- Litigation: within 30 days, the Attorney General must set up an AI Litigation Task Force to challenge state AI laws inconsistent with the order’s policy.
- Evaluation: within 90 days, the Commerce Secretary must publish a review identifying onerous state AI laws, including at minimum laws that require models to alter truthful outputs or compel disclosures that could violate the First Amendment.
- Funding: within 90 days, NTIA must issue a notice making states with such laws ineligible for certain remaining BEAD broadband non-deployment funds, to the extent federal law allows. Agencies must also review whether discretionary grants can be conditioned on state AI policy.
- Regulators: the FCC is to begin a proceeding on a federal AI reporting and disclosure standard that would preempt conflicting state laws, and the FTC must issue a policy statement on how its ban on unfair and deceptive practices applies to AI models.
- Legislation: White House advisers must prepare a recommendation for a uniform federal framework. It is not to propose preempting state laws on child safety, AI compute and data centre infrastructure (other than general permitting reform), or state governments’ own procurement and use of AI.
/ 02Why it matters
For companies operating across many states, a single rulebook would be simpler, and that is the order’s stated aim. Kristian Stout of the International Center for Law & Economics told Route Fifty: “Allowing core technical rules to be fragmented across 50 states” would slow deployment.
But an executive order does not by itself repeal state law. Most of its mechanisms are reviews, lawsuits and funding conditions, each of which takes time and may be contested. For the foreseeable future, organisations will face more uncertainty, not less: state laws on the books, federal challenges pending, and a federal framework that does not yet exist.
It is also worth noting what the order does not touch. The obligations that flow from contracts, sector regulators, existing consumer protection law and courts remain. An AI system that makes a consequential error still produces a liability question, whatever the state of AI-specific statutes.
/ 03The traceability angle
When the rules are unsettled, the safest position is to be able to explain exactly what a system did. Whichever regime eventually applies, federal disclosure standards, surviving state laws, or plain negligence claims, the questions will be similar: what data did the system use, what logic produced the output, and can you show it?
Systems built on a knowledge graph with provenance on every fact, and deterministic reasoning over it, can answer those questions directly. Systems that rely on a model’s opaque internal state can only offer a best reconstruction. That difference does not depend on which level of government wins this argument.
The order’s own focus on “truthful outputs” points the same way. Arguing about whether an output was truthful, or was altered, requires being able to show how it was produced. A system that can display the facts and rules behind an answer is in a far stronger position in that argument than one that can only show the answer.
Preemption changes who writes the rules. It does not change the need to show your work.
/ 04What to watch
Watch for the Justice Department task force in January, and for the Commerce Department’s list of targeted state laws and the FTC policy statement in March. Watch, too, for court challenges to the funding conditions. In the meantime, keep compliance programmes built to the strictest regime you operate under, and invest in the audit trail, which pays off under every scenario.
Whatever the courts decide about preemption, contracts, consumer protection law and sector regulators all remain. The audit question does not go away; it only becomes unclear who will ask it first.
A state benefits agency is a useful test case, since the order keeps state governments’ own procurement and use of AI out of its preemption proposal. If an eligibility recommendation is linked in a graph to the applicant record it read, the statute section and rule version applied, and the caseworker who confirmed it, the agency can answer an appeal, a legislative inquiry or a court challenge from the same record. Deterministic reasoning means a rerun on the same facts yields the same recommendation, so “was this output altered” becomes a comparison rather than a debate. Keeping that graph on the agency’s own infrastructure, behind its firewall, also avoids a second argument about where residents’ data went.
How we’d model it
Eligibility recommendation—readFrom→Applicant recordEligibility recommendation—applies→Statute sectionStatute section—hasVersion→Rule versionEligibility recommendation—confirmedBy→Caseworker
- Ensuring a National Policy Framework for Artificial Intelligence — The White House whitehouse.gov ↗
- Trump signs order targeting 'cumbersome' state AI regulation — Route Fifty (Nextgov/FCW) route-fifty.com ↗
See how Aktiver turns raw data into knowledge graphs your AI can reason over — deterministic, traceable, behind your firewall.