The federal government’s main AI testing shop may be getting a new boss. The Commerce Department plans to move the Center for AI Standards and Innovation, known as CAISI, so that it reports directly to the White House.
Commerce frames the plan as a fix for management and resource problems at the agency.
What CAISI does and what could change
CAISI sits inside the National Institute of Standards and Technology, or NIST, which is part of the Commerce Department. Its job is voluntary testing, evaluation, and standards work for frontier AI models.
That voluntary model has been central to the agency’s work with major developers. CAISI has collaborated with firms including OpenAI, Google, and Microsoft, which gives it a seat at the table in how US AI governance takes shape.
The proposed change would shift who the agency answers to. Under the plan, CAISI officials would report to the White House rather than through the usual Commerce and NIST chain of command.
A rebrand, a revolving door, and a new executive order
CAISI has not always been CAISI. The Trump administration rebranded it in June 2025. Before that, it was the US AI Safety Institute.
Since then, the agency has had trouble keeping anyone in the corner office. Collin Burns departed in April 2026. Chris Fall left shortly after, resigning in July.
No permanent director has been named since Fall’s exit. NIST’s Arvind Raman has been serving as acting director while the turnover continues.
The leadership churn has come alongside a series of interventions that have shaped how the agency operates. Those include the removal of multiple directors and a pause on public assessments of AI models.
There have also been reports that CAISI was sidelined in high-profile decisions, such as export controls.
Then came a June 2026 executive order. It mandated enhanced oversight of CAISI and leaned into voluntary industry reviews.
The order also emphasized classified benchmarking led by national security agencies. At the same time, it limited CAISI’s public outputs and its international collaborations.
Why the reporting line matters
Who an agency reports to shapes what it prioritizes, what it publishes, and how independent its findings appear. Commerce is pitching the move as an operational upgrade. If CAISI truly lacks resources and stable management, a direct line to the White House could, in theory, help it get both.
The flip side is the track record of the past year. Directors have been removed, public model assessments were paused, and an executive order narrowed what the agency can share.
What this means for AI developers and the broader industry
For the companies that work with CAISI, the immediate issue is predictability. Firms like OpenAI, Google, and Microsoft have engaged with the agency on a voluntary basis.
With classified benchmarking led by security agencies, developers may face closer scrutiny of their models. That could influence competitive positioning, especially if some firms are more deeply engaged in those reviews than others.
The reduced public output may matter for the research community too. Published assessments give academics, smaller developers, and international partners a common reference point. If more evaluation work moves into classified channels, that shared baseline gets thinner.
Limits on international collaboration carry their own implications. AI development is global, and standards bodies typically gain influence by coordinating with counterparts abroad.
Several things are worth watching from here. First, whether the reporting shift is formally completed or modified. Second, whether a permanent director is finally named, ending the stretch of acting leadership under Raman. Third, whether public model assessments resume in any form, or whether the classified approach outlined in the June 2026 executive order becomes the default.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.