CAISI director Chris Fall resigns after three months on the job

A federal AI evaluation body losing its director inside a hundred days is the kind of small story that says more about the state of US AI oversight than any hearing. Axios reported on Monday that Chris Fall has resigned as director of the Commerce Department's Center for AI Standards and Innovation, three months after taking the role in April. NIST Director Arvind Raman, who was himself only sworn in on June 30 and previously served as dean of engineering at Purdue University, will oversee CAISI as acting director while Commerce searches for a permanent replacement.

The context makes this messier than a routine departure. CAISI is the reorganized successor to what the previous administration called the US AI Safety Institute, and Fall was not the first choice; Collin Burns, a former researcher at Anthropic and OpenAI, was initially selected but never took the seat. According to the reporting, CAISI has also seen repeated changes in direction, including the quiet removal from its website in May of details about AI testing agreements with Google, Microsoft, and xAI. The office is the government's main touchpoint for evaluating unreleased frontier models from Anthropic, Google's DeepMind, OpenAI, Microsoft, and xAI for capabilities that could aid chemical or biological weapons work, or for training-data corruption.

Why this matters for anyone building or deploying at the frontier: this is the body that will decide what 'tested by the US government' actually means for the next generation of models. Analyst Shashank Joshi put the concern bluntly on X, calling it 'probably the most crucial period for AI safety and testing' while 'the leadership of the main American institution to that end appears to be all over the place'. Labs that had begun to route pre-deployment evaluations through CAISI now face uncertainty on cadence, scope, and who signs off.

The honest caveat is that Commerce says Fall's appointment was always intended to be temporary, that Raman has been reviewing candidates, and that a permanent director is expected in the coming weeks. No reason for the resignation was given, and a spokesperson for the president declined to comment, so anyone building a narrative around policy conflict or private-sector poaching is guessing. The reporting also does not say what happens to those pulled-down testing agreements with Google, Microsoft and xAI, or whether the ongoing frontier evaluations continue on their prior schedule.

For labs, the pragmatic read is to keep investing in their own red-teaming and to expect the US evaluation regime to look less like a settled institution and more like a work in progress well into next year.