The director of the U.S. Center for AI Standards and Innovation (CAISI) stepped in only 3 months before suddenly resigning, with three people changing within six months.

CAISI, the government unit responsible for setting U.S. AI standards and catching cybersecurity risks, has already replaced three heads in a little more than half a year. Its latest director, Chris Fall, only took office three months ago before abruptly resigning; the Department of Commerce did not provide a reason, saying only that the position “was originally meant to be temporary.”

(Background: Part of the Trump administration’s loosening of restrictions on Mythos 5; OpenAI GPT-5.6 Sol is limited to clients approved by the White House.)

(Background add-on: DeepMind CEO argues for establishing an international regulatory body to “review” cutting-edge AI models, which must pass before they can be released to the public.)

Table of contents

Toggle

• Who manages AI risk has been hard to pin down
• The real regulatory big drama—CAISI is often absent
• When China open-source models spark a fight, the regulatory playbook shows

Three months—this is how long Chris Fall has lasted in the director role at CAISI (the U.S. Center for AI Standards and Innovation). On Monday, CAISI confirmed to multiple media outlets that he had resigned, without giving any reason. A Commerce Department official only淡淡 said that Fall’s appointment “was originally meant to be temporary.”

Compared with the previous director, three months is actually already a relatively long stint. Collin Burns took over CAISI in April this year, but he was removed less than a week after taking office. The reason: he had previously worked at Anthropic, and at the time the Trump administration was in a standoff with that company.

Going further back, CAISI’s first person in charge was venture capital insider David Sacks, whose title was “White House AI and Cryptocurrency Czar.” After he stepped down in March, no one has taken over that role to this day. In the past more than half a year, the U.S. government has swapped three people in this AI oversight position; now, it’s NIST Director Arvind Raman acting in the role.

Who manages AI risk has been hard to pin down

CAISI is under the U.S. National Institute of Standards and Technology (NIST). Its predecessor was the U.S. AI Safety Institute from the Biden era; after a reorganization, it became the main federal unit that sets technical standards for AI models, conducts capability testing, and identifies cybersecurity risks. In simple terms, it’s the government’s version of an AI exam unit—at least in theory, any lab publishing a frontier model has to pass its gate first to see whether it’s safe.

Before Fall took over CAISI, he was director of the Office of Science in the first Trump term at the Department of Energy, and also served in an acting capacity at the Department of Energy’s Advanced Research Projects Agency–Energy (ARPA-E). Earlier still, he worked at the Office of Naval Research. With a full resume, he still couldn’t stick around.

The Department of Commerce said it would announce the official successor “within the coming weeks.” But from Sacks to Burns to Fall, that line sounds like it’s being replayed.

The real regulatory big drama—CAISI is often absent

The irony is that in the past half year, CAISI was nearly absent from the most high-profile AI oversight issues. In June, the U.S. Department of Commerce cited a little-known export-control provision that temporarily forced Anthropic to take its Mythos and Fable models offline. The one who moved was Commerce Secretary Howard Lutnick, not CAISI. By the end of the month, Lutnick said he was satisfied with Anthropic’s security plan, the ban was lifted, and the whole back-and-forth likewise had no sign of CAISI.

Earlier this month, the White House signed an executive order launching a cybersecurity vulnerability coordination mechanism called “Gold Eagle,” naming the Commerce Department and the Department of Homeland Security as responsible parties. CAISI’s name also did not appear on the list.

At the same time, after Anthropic’s model unfreeze, Google DeepMind CEO Demis Hassabis began calling for the industry to emulate FINRA, the securities industry self-regulatory organization, and establish an independent frontier AI standards body led by the industry rather than the government—which is exactly the mission CAISI was originally set up to fulfill.

When China open-source models spark a fight, the regulatory playbook shows

On the weekend before Fall resigned, the Kimi open-source model’s new version released by Moonshot showed benchmark results catching up to mainstream frontier models, prompting debate inside the government again about whether to shut down China’s open-source models. In this dispute, Sacks also jumped in to rebut, arguing that regulation should not be used as a trade tool to protect U.S. private model vendors.

And there’s not much CAISI can point to in this debate. It has indeed issued several assessment reports on China open-source weight models (such as Z.ai’s GLM-5.2 and DeepSeek V4 Pro), but how the testing process itself runs and how standards are set has never been publicly explained.