Does the US have a federal AI safety law? No. As of September 2026 there's no comprehensive federal statute on AI safety — oversight runs on executive orders, agency enforcement, and bills that never made it out of Congress, while more than 100 state laws cover the gap. That split is what Washington is fighting over right now.
I spent an evening trying to work out which rules would apply to one AI product sold in three different states, and I gave up somewhere around the fourth browser tab. So here's the version I wish I'd found: what actually exists, why Congress keeps stalling, and what changed in the last few days.
Split graphic contrasting an empty federal shelf labeled "Federal: no comprehensive AI safety law" with a stack of state law books labeled "States: 109 AI laws, as of July 1, 2026". AI-generated image
Does the US have a federal AI safety law? What exists instead
Does the US have a federal AI safety law? Not in the way most people assume. Federal AI governance in September 2026 runs on three thinner sources — presidential executive orders, enforcement by existing agencies under existing statutes, and bills still sitting in committee — as laid out in the Congressional Research Service's AI policy report.
The loudest of those sources is an executive order. On December 11, 2025, President Trump signed "Ensuring a National Policy Framework for Artificial Intelligence," which pushes to preempt state AI laws rather than write a federal safety standard. Law firm analyses from Gibson Dunn and Ropes & Gray read it the same way.
It came with teeth. The order stood up an AI Litigation Task Force inside the Justice Department, operating from January 10, 2026, whose job is to challenge state AI laws in federal court.
Then came the paperwork. The White House released a National AI Policy Framework on March 20, 2026 with seven policy pillars — child safety and state-law preemption among them — recommended to Congress. Two days earlier, Senator Marsha Blackburn of Tennessee had published a 291-page discussion draft called the TRUMP AMERICA AI Act, which would put a duty of care on AI developers and turn the preemption order into statute.
None of that is a law yet. A framework is a wish list, and a discussion draft is a starting gun.
Why hasn't Congress passed a federal AI safety law?
Congress hasn't passed one because two disagreements keep colliding: whether states or Washington should hold the power to regulate AI, and whether safety rules would hand China the lead in the meantime. Bipartisan working groups produced no bill, and the Senate's clearest vote on the question went 99-1 against overriding state law.
That vote is worth sitting with. On July 1, 2025, the Senate stripped a 10-year moratorium on state AI laws out of the budget reconciliation bill, 99 to 1.
Senators who agree on almost nothing agreed on that. A version of the moratorium was also left out of the 2026 defense authorization bill, according to reporting by StateScoop, though I haven't seen that confirmed elsewhere.
So the gridlock isn't apathy. Blocking a federal override turns out to be much easier than agreeing on what a federal AI safety law should say.
The states moved first, and Washington wants it back
State legislatures filled the vacuum fast. By TechPolicy.Press's count, the 50 states had enacted at least 109 AI-related laws and 28 data center laws as of July 1, 2026.
California went furthest: Governor Gavin Newsom signed SB 53, the Transparency in Frontier AI Act, on September 29, 2025 — the first US state law aimed squarely at frontier model safety.
SB 53 is narrow on purpose. It targets "large frontier developers," meaning companies above $500 million in annual revenue that train models using more than 10^26 FLOPs of compute, and it requires them to publish a safety framework, report safety incidents, and protect whistleblowers. Civil penalties run up to $1 million per violation.
| Federal | State | |
|---|---|---|
| Comprehensive AI safety law | None — executive orders, agency enforcement, pending bills | Yes — at least 109 AI laws (as of July 1, 2026) |
| Signature move | Trump preemption order (Dec 11, 2025); White House framework (Mar 20, 2026) | California SB 53, signed Sep 29, 2025 |
| Latest legislative result | Senate killed the moratorium 99-1 (Jul 1, 2025) | State-by-state lawmaking still expanding |
My honest read: the patchwork is a mess for companies and better than nothing for everyone else. If you want one national standard, you still have to name what the standard says — and nobody in Washington has.
Where the fight stands in September 2026
The week of September 11–13, 2026 put every side on the record at once. House Speaker Mike Johnson said Congress won't lead on AI safety regulation because the US could fall behind China, President Trump rejected calls to slow AI development for the same reason, and former President Obama pushed Democrats to make AI safety a central midterm issue.
Johnson did leave a door open. He told Axios he'd bring the House back to vote "if there's a solution" — which reads less like a plan than a dare.
The industry made it messier. On September 12–13, Anthropic CEO Dario Amodei warned that AI could slip out of human control and urged competitors to slow down; Altman and Musk publicly agreed with him. Altman also ruled out an OpenAI IPO in 2026, saying it's not the right moment, and said an extinction-level risk is unacceptable even at very low odds.
Set that against Johnson's position and you get the oddest scoreboard of 2026: AI executives asking for restraint, congressional leadership declining to legislate it.
Horizontal timeline of US AI policy from July 1, 2025 Senate 99-1 moratorium vote through Sep 29, 2025 California SB 53, Dec 11, 2025 Trump preemption order, Mar 2026 Blackburn draft and White House framework, to Sep 13, 2026 statements by Johnson, Trump, Obama and AI CEOsAt a glance
- Who this hits: AI companies juggling state-by-state compliance, voters heading into the November midterms, and users whose protections depend on their state.
- Where it stands: as of September 2026, no federal AI safety law; the Senate killed the state-law moratorium 99-1; California's SB 53 is the template other states copy.
- The number to know: up to $1 million per violation under SB 53 — in California only, and only for the largest frontier developers.
- One caution: this moves fast. The answer to does the US have a federal AI safety law could flip on a single bill or a single preemption ruling.