The U.S. government's main AI testing agency has now cycled through three leaders in about six months. Chris Fall resigned today as director of the Center for AI Standards and Innovation, known as CAISI. Three months in the job, no stated reason, and the Commerce Department's answer was essentially that he was never meant to stay long.
This is not the first time. Before Fall, Collin Burns lasted four days before being pushed out, reportedly because he had previously worked at Anthropic, a company the administration had been fighting with. Before Burns, the role was occupied by venture capitalist David Sacks in his capacity as White House AI and crypto czar. He left in March. The NIST director will run CAISI in the interim.
What CAISI actually does matters here. It is the government body that tests powerful AI systems before companies release them to the public. The testing focuses on specific risks: whether a model could help someone plan a cyberattack, assist in building a biological weapon, or be exploited by foreign adversaries. Every major AI company, including Google, Microsoft, OpenAI, Anthropic, and Elon Musk's xAI, has a voluntary agreement with CAISI to hand over their newest models for this kind of scrutiny before public release. That scrutiny is what gives governments, businesses, and regulators some assurance that these tools are not carrying hidden dangers.
For business operators, that context matters for a practical reason. If you use AI software from any of these major providers, the pre-deployment testing that CAISI runs is part of the supply chain of trust behind those products. An agency with unstable leadership, shrinking visibility inside its own government, and unanswered questions about its processes is a weaker link in that chain.
The instability at the top is only part of the story. The agency is showing signs of being edged out of the White House's broader AI security agenda. When the administration launched Gold Eagle, a new cybersecurity program created specifically to coordinate AI-related threats, CAISI was not named among the federal bodies involved. That is a pointed omission for an agency whose entire mandate is AI security testing and standards. The Gold Eagle program is run out of Treasury, Homeland Security, and Defense.
Meanwhile, in May, the Commerce Department announced that Google, Microsoft, and xAI had signed formal pre-deployment testing agreements with CAISI. Seven days later, the page describing those agreements was deleted from the government website with no explanation. The testing program appears to still be running, but the public record of it vanished.
Separately, CAISI has published a handful of reports on Chinese AI models, including evaluations of systems from Z.ai and DeepSeek. But it has declined to explain how those evaluations work, and multiple press inquiries have gone unanswered since early July.
All of this adds up to an agency that is nominally responsible for a critical function but is operating without stable leadership, with reduced visibility in White House planning, and with less transparency about its own methods than at any point since it was created.
For businesses that depend on AI tools for operations, procurement, or customer service, the practical takeaway is straightforward: the government body that is supposed to independently verify whether powerful AI is safe to use is going through an extended period of instability. That does not mean the products you are using are unsafe. It does mean the independent verification layer is thinner right now than it should be. Paying attention to what the major AI labs publish about their own safety testing, and asking vendors direct questions about how their models are evaluated, is a reasonable response to that gap.