The push for Federal AI standards is back on, with a Truth Social post by the President, followed by a drumbeat from the administration and tech allies.

The policy failed to pass in the OBBBA, but it's good policy and should pass this time around.

According to the National Conference of State Legislatures, 635 AI-related bills were introduced across US states in 2024, with 99 enacted. In 2024 alone, at least 45 states plus several territories introduced AI bills, and 31 states enacted laws or resolutions. In 2025 the pace accelerated: all 50 states, Puerto Rico, the Virgin Islands and Washington, DC introduced AI measures, with roughly 100 new laws or resolutions adopted by mid-year. A recent US Chamber analysis estimates that more than 1 100 AI-related bills were filed in statehouses in 2025 alone. 

This is madness. We can't regulate a transformative technology like AI at the state level.

More specifically, California is of course a great boogeyman. SB 1047, the "Safe and Secure Innovation for Frontier Artificial Intelligence Models Act," was the first serious attempt anywhere to build a bespoke regime for so-called "frontier" models. It targeted systems above a very high compute threshold (training runs on the order of 100 million dollars), and would have imposed ex ante safety plans, incident reporting, and shutdown mechanisms, backed by enforcement from the Attorney General.  Governor Newsom ultimately vetoed the bill in September 2024, but the basic architecture has already shaped subsequent proposals in California and beyond. 

AB 2930, aimed at "automated decision tools," goes further downstream. It obliges developers and deployers of tools used in "consequential decisions" (credit, employment, housing, access to benefits) to conduct detailed impact assessments, maintain governance programmes to manage "algorithmic discrimination," and provide documentation to regulators. Violations involving algorithmic discrimination can trigger civil penalties up to 25 000 dollars per violation, enforceable by the Civil Rights Department and other public authorities.  This is a full-dress compliance regime for almost any serious use of machine learning in business.

Since then California has enacted SB 53, the "Transparency in Frontier Artificial Intelligence Act," which requires large model developers to publish safety protocols, report serious incidents and meet emerging national and international standards, again with thresholds tied to massive training runs. Taken together, these measures amount to a prototype of a vertically integrated state AI code, with extraterritorial reach to any firm doing business in California.

And, of course, we have to say it. Chyna. We are in an AI race with China, which we need to win.

Beijing has had, since 2017, a unified "New Generation Artificial Intelligence Development Plan" whose explicit goal is for China to become the world’s primary AI power by 2030.

At the federal level, the Trump administration’s “America’s AI Action Plan” frames AI dominance as a central objective for economic, military and diplomatic power, and calls for removing barriers to AI infrastructure and exports while maintaining core safety and civil-liberty protections. At the state level, however, we see highly heterogeneous regulatory experiments, from narrow rules on deepfakes to expansive codes like Colorado’s AI Act and California’s automated decision tools regime.

We need a unified Federal approach and a moratorium on state AI bills. At the risk of using a hoary cliché: it's just common sense.

Policy News You Need To Know

#VoxClamantisInDeserto — The spectacle of the House vote on the release of the so-called Epstein Files was sad and embarrassing. Which is why we want to single out for praise the lone voice speaking in the desert, Republican Congressman Clay Higgins, the lone "No" vote, and quote his justification in extenso, which is entirely correct: "I have been a principled “NO” on this bill from the beginning. What was wrong with the bill three months ago is still wrong today. It abandons 250 years of criminal justice procedure in America. As written, this bill reveals and injures thousands of innocent people – witnesses, people who provided alibis, family members, etc. If enacted in its current form, this type of broad reveal of criminal investigative files, released to a rabid media, will absolutely result in innocent people being hurt. Not by my vote. The Oversight Committee is conducting a thorough investigation that has already released well over 60,000 pages of documents from the Epstein case. That effort will continue in a manner that provides all due protections for innocent Americans. If the Senate amends the bill to properly address privacy of victims and other Americans, who are named but not criminally implicated, then I will vote for that bill when it comes back to the House." Quite right.

#Nukes — DOE has approved a roughly $1 billion facility through LPO for Constellation Energy to restart Three Mile Island’s Unit 1 by 2027, restoring nearly 1 GW of firm nuclear capacity to the PJM grid with a parent guarantee to shield taxpayers. Beyond the symbolism, it's a good sign that the admin is putting real support behind nuclear.

#Idiocracy — The Atlantic did an article on that UCSD report we mentioned earlier. The report finds that more than 900 incoming freshmen now arrive with math skills below high-school—and often even middle-school—level. The university is now running de facto primary-school remediation even for students who earned straight A grades and "advanced" math credits in high school. This collapse in quantitative competence mirrors national data: eighth-grade math achievement has fallen back by roughly a full school year from its 2013 peak, erasing half a century of gains, with the steepest drops among weaker students. Policy choices are squarely in the dock: the federal retreat from Bush-era test-based accountability after 2015, district-level grade inflation and "no zeros" rules, pandemic-era school closures that devastated learning while teachers’ unions blocked reopening, and ideological moves such as the UC system’s elimination of SAT/ACT requirements in the name of "equity," which also removed the most reliable signal of math readiness. For Republican lawmakers and reformers, the agenda almost writes itself: restore rigorous, transparent accountability for K–12 systems; re-establish objective admissions testing, especially in math-intensive tracks; protect access to advanced courses rather than diluting them; and treat foundational numeracy as a core national competitiveness and security priority.

#RuleOfLaw — A $16 million fraud network has been uncovered and prosecuted in California’s hospice program. Much like the autism fraud network discovered in Minneapolis, and the adult day care fraud network in Brooklyn, the scam operated by stealing the identities of foreign nationals in order to charge medicaid for phony services. The hospice facility was opened and operated by using these stolen identities and the money was laundered through a series of shell companies also opened in their names. There is a nationwide trend of immigrant fraud networks whose activities went unnoticed by the Biden administration's less-than-rigorous enforcement, now being uncovered and prosecuted by the Trump administration.

#RareEarths — A mine in Alaska has been found to contain enormous amounts of rare earth minerals, potentially securing a new domestic source to a crucial component in American energy independence. We shouldn't pop the champagne yet. Getting those crucial rare earths involves not just digging them out of the ground, but refining them, which requires overhauling our regulatory environment and rebuilding very elaborate process knowledge that China has monopolized. Still, good news!