Trump's AI safety agency loses its third leader in under a year

NIST Loses Third AI Standards Director as U.S. Safety Framework Drifts

Chris Fall has resigned as director of the Center for AI Standards and Innovation after just three months on the job 1. NIST Director Arvind Raman will step in as acting head, a Commerce Department spokesperson confirmed 1. No reason was given for the departure 2.

This is not a Washington personnel shuffle. CAISI is the lead U.S. body charged with writing technical standards and evaluation methods for AI systems and gauging their cybersecurity exposure 2. If you build AI systems in the United States, this is the body that would define what "tested" and "safe" mean before a model reaches the public. It has now lost three leaders since March.

David Sacks held the White House AI and crypto czar role before stepping down in March; he has not been replaced 1. Collin Burns took the CAISI directorship next and lasted less than a week. He was reportedly pushed out in April because he had previously worked for Anthropic, a company the Trump administration was then clashing with 2. Fall, a former director of the Department of Energy's Office of Science during the first Trump administration, was appointed in April 2. He made it to July.

Three leaders out in under five months is not a string of bad hires. It is a structural signal.

When the White House announced its new Gold Eagle cybersecurity clearinghouse earlier this month, CAISI was not among the federal agencies named to participate, even as the Commerce Department and Department of Homeland Security were both included 2. When the administration sidelines the body you are supposed to lead, retaining the job becomes difficult. Separately, TechCrunch reports sending multiple inquiries to the Commerce Department and NIST since July 9 asking how their large language model evaluations actually work, with no response received 2.

The opacity and the churn feed each other. If the director keeps changing, no one owns the testing methodology. If no one owns the methodology, companies cannot predict what the standards will look like. And the global regulatory landscape is not waiting.

The EU AI Act is now enforcing. China is publishing its own AI governance frameworks. The U.S. body chartered to produce a credible, technically grounded alternative cannot retain a director for a single quarter 1.

That vacuum has real costs. In June, OpenAI agreed to restrict its GPT-5.6 model to a group of trusted partners at the request of the U.S. government 1. Two weeks earlier, the Commerce Department invoked an export control directive that forced Anthropic to pull its Fable 5 and Mythos 5 models from the market entirely 1. Both companies eventually released their models more broadly, but the stop-and-go process exposed a system where the rules are being decided case by case rather than governed by durable standards 1.

For investors, the pattern is the data point. When the EU passes a regulation, companies know what to build toward. When China releases a framework, companies know what to comply with. When the U.S. standards body churns through leaders every few weeks, companies architect toward uncertainty, and uncertainty carries a price tag.

Industry leaders are already looking past CAISI. Google DeepMind CEO Demis Hassabis has started pushing for an independent, industry-run organization to set frontier AI standards, pointing to FINRA, the financial industry's self-regulatory body, as a model 2. That is essentially the mission CAISI was created to handle. When the head of one of the world's most advanced AI labs suggests building a parallel institution to do your job, the vote of no confidence is clear.

CAISI has produced some work despite the instability. It has published assessments of Chinese open-weight models including Z.ai's GLM-5.2 and DeepSeek V4 Pro 2. But it has not publicly disclosed how its model evaluations are conducted 2.

Fall's resignation landed the same weekend that Chinese AI lab Moonshot released Kimi K3, which the company says matches or surpasses leading U.S. models on some benchmarks 1. The administration is reportedly weighing restrictions on Chinese open models, a debate that flared over the weekend 2. Sacks himself argued against using regulations as a protectionist shield for U.S. proprietary labs 2.

The U.S. is making consequential decisions about AI governance with a standards body that has seen three leaders walk away since spring. The pattern is the story.