Trump’s latest AI czar has already resigned

Chris Fall lasted three months. Collin Burns lasted a week. David Sacks lasted until March. The Center for AI Standards and Innovation has become a revolving door for appointees the administration cannot keep or does not want. CAISI sits inside NIST with a mandate to develop technical standards, testing methods, and cybersecurity risk assessments for AI models. That is the mission. The reality is different.

When the Commerce Department invoked an obscure export directive to yank Anthropic's Mythos and Fable models off the market in June, CAISI was not in the room. Secretary Lutnick lifted the ban weeks later, satisfied with Anthropic's safety plans. CAISI was not in that room either. Then the White House signed the Gold Eagle executive order, creating a clearinghouse for cybersecurity vulnerability coordination across federal agencies. The Commerce Department and Homeland Security were named. CAISI was not.

An agency designed to lead on standards and testing has been sidelined by its own administration. The pattern is not accidental. Burns was pushed out because he came from Anthropic while the administration was fighting Anthropic. Fall brought Department of Energy pedigree but no visible authority. Sacks, a venture capitalist, held the title of White House AI and crypto czar before stepping down. None of them stayed long enough to build institutional memory, let alone institutional power.

Meanwhile, the frontier moves elsewhere. Demis Hassabis, CEO of Google DeepMind, now calls for an independent, industry-run standards body modeled on FINRA — a self-regulatory organization with teeth. That is exactly what CAISI was supposed to be. The fact that the industry's most prominent research leader thinks a new body is needed says everything about CAISI's credibility vacuum.

The administration's response to Chinese open models confirms the disorder. Moonshot's Kimi model posted competitive benchmarks against flagship U.S. systems. Axios reported the administration weighed banning Chinese open models outright. Sacks, the former czar, immediately countered that regulation should not become protectionism for U.S. proprietary labs. The debate erupted over a weekend. CAISI, which has published a few reports on Chinese models like Z.ai's GLM-5.2 and DeepSeek V4 Pro, has said little about how it tests them or what standards it applies. Its processes remain opaque. Its voice is absent.

Open-weight models change the calculus. They can be downloaded, inspected, modified, deployed without permission. Banning them at the border is not like blocking a closed API. It requires a standards framework that can evaluate capability, risk, and provenance transparently. CAISI was built for that framework. Instead, the administration reaches for export controls and ad hoc bans — tools designed for hardware and encryption, not for model weights that replicate across the internet in hours.

Gold Eagle promises a clearinghouse. But a clearinghouse without the standards body that defines what gets cleared is a warehouse without a manifest. The executive order names coordination. It does not name authority. It does not name CAISI.

The revolving door at CAISI is not a personnel problem. It is a structural signal. The administration does not know whether it wants a standards setter, a gatekeeper, or a political instrument. It has tried all three and kept none. Burns was too close to industry. Fall was too bureaucratic. Sacks was too political. The next appointee will face the same contradiction: an agency with a technical mandate but no policy seat at the table where decisions actually get made.

Hassabis's FINRA analogy is precise. Financial regulation works because the industry funds and submits to a body that can enforce rules. AI has no such body. The U.S. government could have built one inside CAISI. It chose instead to marginalize it, bypass it, and now watch its leaders quit in succession.

The Chinese model challenge is not waiting. Open weights from Moonshot, DeepSeek, Z.ai, and others will keep arriving. They will be benchmarked, forked, integrated into products worldwide. The U.S. needs a standards regime that can assess them on merit and risk — not on nationality. CAISI could be that regime. It has the statutory mandate. It has the NIST infrastructure. It lacks the political backing to act.

Fall's resignation is a symptom. The disease is a strategy vacuum. The administration invokes export controls on Monday, signs executive orders on Tuesday, floats protectionist bans on Wednesday, and leaves its standards agency leaderless on Thursday. The rest of the world — allies, adversaries, open-source communities — watches and codes.

If the U.S. wants to lead on AI safety, it must empower the body it created to lead. If it wants to protect domestic labs, it should say so and accept the innovation cost. What it cannot do is maintain a zombie agency, cycle through directors, and pretend standards emerge from chaos. They do not. They emerge from authority, continuity, and the willingness to let technical judgment override political impulse. CAISI has none of the three.