President Donald Trump has signed a new executive order on artificial intelligence that rescinds or rewrites major pieces of the Biden administration’s 2023 AI directive and tells federal agencies to “remove barriers” to AI development. The move shifts the federal posture from “guardrails first” to “growth first,” setting up clashes with states that passed or proposed their own AI safety rules and kicking key questions back to Congress and the courts.
Here’s what the order does, what disappears with Biden’s directive revoked, and what companies, workers, and consumers should expect next.
What the order does
The White House framed the action as “Removing Barriers to American Leadership in Artificial Intelligence,” instructing agencies to streamline approvals, scrub rules seen as obstacles to AI deployment, and use AI more aggressively in federal operations. In practical terms, it means:
- Agencies must review existing regulations and guidance that could slow AI adoption and propose revisions.
- Federal procurement is encouraged to pilot AI tools where they can reduce cost and improve service.
- New agency rules specific to AI face a higher bar: cost‑benefit tests and coordination with the Office of Management and Budget to avoid “duplicative” oversight.
- The order signals federal preemption in areas where state AI controls may conflict with federal programs, a flash point for looming legal fights.
At the same time, the order revokes the Biden administration’s 2023 AI executive order, which had required safety testing and reporting for the largest models, set deadlines for watermarking standards and red‑team evaluations, and pushed agencies to assess risks in critical sectors before deploying AI.
Reuters reported the revocation; the White House published the new directive’s text.
What disappears with Biden’s order
Biden’s 2023 executive order was sprawling. It directed Commerce, NIST, Homeland Security, and others to create new testing practices and reporting regimes for “frontier” models—systems trained at massive compute levels—and to develop content provenance tools to flag AI‑generated media. With Trump’s order rescinding that framework:
- Model reporting to Commerce halts unless Congress enacts a statutory requirement or an agency re‑imposes it under new authority.
- Timelines for watermarking or content authenticity standards become voluntary. NIST can still publish frameworks, but agencies are no longer told to require them.
- Sector risk assessments—e.g., for critical infrastructure, health, and finance—are now at agency discretion, not a White House‑set mandate.
In short, the new posture leans on existing, general laws (privacy, consumer protection, civil rights, competition) rather than AI‑specific obligations.
What changes immediately
Three shifts are visible right away.
First, agencies will green‑light more pilots. Expect AI in call centers, benefits processing, fraud detection, and IT help desks to roll out faster. Procurement officers have clearer cover to try tools and learn on the job.
Second, compliance calendars shrink. Developers working toward Biden‑era reporting thresholds, model incident reporting, or watermark deadlines will downshift to “monitor and wait.” Voluntary frameworks (NIST’s AI Risk Management Framework, for example) remain, but the federal hammer is gone.
Third, legal preemption debates begin. The order signals that federal standards should trump state rules in areas such as federal contracting, interstate markets, or national security. States with active or proposed AI laws—California, Colorado, and others—will test those boundaries in court.
For Instance:
A developer training a large language model this spring had mapped to the Biden order’s reporting triggers (compute thresholds, safety testing summaries to Commerce). With the new order, they can pause those reports and rely on internal safety processes and voluntary evaluations—unless a regulator (FTC, SEC, banking agencies) exercises existing authority in their sector.
What does not change
The new order does not repeal the Federal Trade Commission Act, civil rights laws, antidiscrimination statutes, copyright laws, or product liability standards. Nor does it insulate companies from enforcement if their AI systems deceive consumers, discriminate, or create unsafe products.
Key guardrails that remain:
- The FTC can pursue “unfair or deceptive” AI claims (e.g., deepfake ads, exaggerated abilities).
- EEOC and DOJ retain authority over employment discrimination, including biased screening tools.
- Copyright rules still govern training data, outputs, and licensing; courts continue to resolve disputes.
- Existing cybersecurity directives apply to AI‑enabled systems used in critical infrastructure.
NIST and other standards bodies can keep publishing best practices, and agencies can still adopt them. The difference is compulsion: it’s now “can,” not “must.”
Industry and civil liberties reactions
Technology companies, startups, and some business groups welcomed the shift as a chance to accelerate responsible deployment without red tape. They argue that fast‑moving fields need flexible rules and that existing laws can handle bad actors.
Civil liberties and labor advocates criticized the order for weakening transparency and sidelining protections, especially for deepfakes, biometric surveillance, hiring tools, and models used in critical infrastructure. They warned that preempting state rules could block experiments with more protective approaches.
Both sides agree on one thing: Congress remains the missing piece. Without legislation, executive orders can swing with administrations.
Federal vs. state: Where fights will land
Expect three arenas for the next round:
- Procurement and programs. If a state AI rule conflicts with federal requirements or frustrates federal contracting, the administration will argue preemption. States will test the limits.
- Commerce Clause. For interstate AI services (cloud‑hosted models, nationwide APIs), the administration may claim national standards should control to avoid burdens on interstate commerce.
- Civil rights and consumer protection. States will argue their AI rules complement, not conflict with, federal law. Expect coordinated state AG actions in deceptive deepfake cases and data privacy.
For Example
A state passes a law requiring “frontier” model developers above a compute threshold to perform third‑party safety audits. A developer serving a federal agency claims the state rule is preempted for that deployment. The question lands in court: does the federal policy space foreclose stronger state rules in this context?
What companies should do now
Even with fewer mandates, the risk calculus did not disappear. Practical steps:
- Keep internal safety testing. Red‑team high‑risk systems and document mitigations; it’s cheaper than discovery after a breach or scandal.
- Use content provenance where it matters. Watermarks and signed assets build trust, even if voluntary.
- Map sector regulators. If you touch finance, health, education, housing, or kids’ privacy, your obligations come from existing laws, not the new AI order.
- Watch states. If you operate nationally, design for the strictest applicable state rule rather than whack‑a‑mole exceptions.
- Train humans in the loop. Many failures are operational (over‑reliance, poor review) rather than model flaws.
To give you an example: A bank adopting an AI underwriting tool should keep a human‑in‑the‑loop review, test for disparate impact, and prepare an adverse action explanation process. None of those depends on an AI‑specific mandate; they flow from fair lending laws and good governance.
What citizens will notice
You will see more AI features in government services: chat for common questions, fraud screening, faster checks on applications. You will also see more AI in private‑sector tools without new federal disclosure requirements: AI in hiring, customer service, and advertising, including deepfakes.
If your state has passed an AI transparency or safety law, you may see different disclosures or options than your friend in another state. The patchwork is back.
What to watch next
- Agency guidance. OMB, Commerce, and NIST will translate the order into memo‑speak. Procurement rules and pilot programs will follow.
- Congress. Narrow bills (deepfake election ads, data portability, kids’ safety) are likelier than a sweeping AI act. Watch for bipartisan carve‑outs.
- Courts. State‑federal preemption challenges and copyright cases over training data will define the edges of the new landscape.
- Election ads. As campaign season heats up, deepfake enforcement will test both federal and state tools without a Biden‑style national watermark rule.
Bottom line
The federal government just changed lanes on AI policy. The Trump administration’s order removes Biden‑era hurdles and invites agencies to adopt and promote AI quickly. It also bets that existing laws and voluntary standards are enough to police harm—while warning states not to get in the way.
That means more AI, sooner, especially in government services and heavily regulated industries that can point to compliance playbooks they already know. It also means uneven protection: strict in some states, lighter in others, with disputes migrating to courtrooms and committee rooms. Until Congress writes durable guardrails, executive orders will keep steering—and the wheel will feel loose every four years.
