Hugging Face has thrown its weight behind the National AI Research Resource, but with a clear message: the White House’s roadmap needs more than compute and data. In a late June submission to the Office of Science and Technology Policy and the National Science Foundation, the company laid out five recommendations that read less like a wish list and more like a governance blueprint. The memo, which responds to the NAIRR Task Force’s interim report, urges the appointment of technical and ethical experts as advisors — not as a courtesy, but as a calibration mechanism for what’s feasible and what’s harmful.
Named directly in the filing: Dr. Margaret Mitchell, Hugging Face’s Chief Ethics Scientist and one of the most prominent figures in AI ethics after her high-profile departure from Google. The company positions her as a ‘natural example’ of the kind of external advisor NAIRR needs. That’s a pointed suggestion, given Mitchell’s history of speaking out about bias in large language models and the institutional failures that allowed them to ship unchecked.
The recommendations go beyond personnel. Hugging Face wants NAIRR to establish documentation standards, pointing to Model Cards as a template that’s already widely adopted. It wants low- and no-code tools — citing its own AutoTrain product — so that interdisciplinary researchers without engineering backgrounds can actually use the resource. It wants monitoring for open-source models with high misuse potential, including Responsible AI Licenses that would give NAIRR legal teeth if someone abuses the platform. And it wants resources available in multiple languages, citing the BigScience Research Workshop, a 1,000-plus researcher collaboration with the French government that produced one of the most powerful open multilingual models to date.
What’s striking here is the timing. The NAIRR is still largely theoretical — a proposed national infrastructure that would democratize access to compute, data, and models. But Hugging Face is treating it as an imminent reality, with concrete, addressable failure modes. The company’s framing — ‘harm must be defined by NAIRR and advisors and continually updated’ — reflects the hard lesson of the last few years: that bias, disinformation, and hate speech aren’t edge cases. They’re default behaviors when systems are built without guardrails.
The subtext is hard to miss. Open-source AI has been under scrutiny, with critics arguing that releasing powerful models without restrictions invites misuse. Hugging Face is essentially saying: open is fine, but only if the infrastructure around it is deliberate. The question now is whether the White House sees these recommendations as a partner’s offer of expertise or a vendor’s pitch for relevance. Either way, the NAIRR conversation just got more specific.
