US AI Safety Chief Chris Fall Resigns After 3 Months

A Commerce Department spokesperson confirmed Fall’s resignation without disclosing the reason, according to the report. CAISI is part of the Commerce Department.
The director of the National Institute of Standards and Technology (NIST), Arvind Raman, will serve as acting director of CAISI while continuing to oversee NIST, the report said, citing the spokesperson.
The report said that when the White House announced a clearinghouse called Gold Eagle that will focus on finding and fixing cyber vulnerabilities, CAISI was not named as an agency involved with its development.
Fall’s departure leaves the White House without a clear pointperson for AI policy and research. Former White House AI and crypto czar David Sacks stepped down from that role in March and has not been replaced, the report said.
The Hill reported Monday that Fall’s predecessor at CAISI, Collin Burns, was pushed out of that position by the White House four days after he started due to his ties to Anthropic, which clashed with the administration.
The report said Fall’s position was always meant to be temporary.
Reuters reported Monday that CAISI is responsible for working with leading AI labs such as Anthropic, Google DeepMind and OpenAI, testing their unreleased models for vulnerabilities, calculating risks posed by advanced models and limiting the opportunities for U.S. adversaries use AI.
PYMNTS reported in May that Google DeepMind, Microsoft and xAI signed agreements with CAISI to share their frontier AI models with the agency for national security testing before releasing those models to the public. The companies joined Anthropic and OpenAI, which signed agreements with CAISI’s predecessor in August 2024 and later renegotiated these agreements to match CAISI’s directives.
In June, Trump signed an executive order that asks companies to provide the White House with access to new AI models before their release, take part in benchmarking to examine each model’s capabilities, and allow the government’s trusted partners access to models to strengthen the cybersecurity of critical infrastructure.

Facts Only

* A Commerce Department spokesperson confirmed Fall’s resignation without disclosing the reason.
* CAISI is part of the Commerce Department.
* Arvind Raman, director of NIST, will serve as acting director of CAISI and continue to oversee NIST.
* CAISI was not named as an agency involved in the development of the Gold Eagle clearinghouse announced by the White House.
* Fall’s departure leaves the White House without a clear pointperson for AI policy and research.
* Collin Burns, Fall’s predecessor at CAISI, was pushed out four days after starting due to ties to Anthropic.
* CAISI is responsible for working with leading AI labs like Anthropic, Google DeepMind, and OpenAI to test unreleased models for vulnerabilities.
* Google DeepMind, Microsoft, and xAI signed agreements with CAISI to share frontier AI models for national security testing.
* Companies joined Anthropic and OpenAI, who signed agreements with CAISI's predecessor in August 2024.
* Trump signed an executive order asking companies to provide access to new AI models, participate in benchmarking, and allow government partners access to models for critical infrastructure cybersecurity.

Executive Summary

A Commerce Department spokesperson confirmed Fall's resignation without providing a reason, and CAISI is part of the Commerce Department. Arvind Raman, director of NIST, will serve as acting director of CAISI while continuing to oversee NIST. The White House announced a clearinghouse called Gold Eagle for finding and fixing cyber vulnerabilities; CAISI was not named in this development. Fall's departure leaves the White House without a clear pointperson for AI policy and research, following the stepping down of David Sacks from that role in March. Collin Burns, Fall's predecessor at CAISI, was removed four days after starting due to ties to Anthropic. CAISI is responsible for testing unreleased models from labs like Anthropic, Google DeepMind, and OpenAI for vulnerabilities and calculating risks posed by advanced models. Agreements exist where these AI labs shared frontier models with CAISI's predecessor before releasing them publicly. An executive order was signed by Trump requiring companies to provide access to new AI models, participate in benchmarking, and allow government partners access to models to enhance critical infrastructure cybersecurity.

Full Take

The narrative centers on the institutional friction between administrative restructuring and the complex technological demands of AI governance. The stated goal appears to be a focus on national security and vulnerability testing related to frontier AI models through an established agency, CAISI. However, the reported internal shifts—such as replacing a director due to external alignment conflicts with the administration—suggest that the mechanisms for policy direction are subject to political influence rather than purely technical mandate. The history of agreements involving major labs (Google DeepMind, OpenAI) and the operational role of CAISI imply a system built on cooperative risk mitigation, yet the subsequent personnel changes indicate instability in who controls that coordination. Furthermore, the executive order demanding access and benchmarking suggests an explicit tension between open-source development by AI entities and government security imperatives. The pattern here is a tension between mandated public-facing policy (the White House announcements) and the internal, high-stakes operational reality of advanced technology oversight. What is not fully explored is how this dynamic affects the long-term coherence of cross-sector partnerships when leadership positions are volatile.