Updated 2026-10-25
from the news desk
Should AI research follow human subjects ethics rules? NIST still lists Belmont AI project
NIST’s Human-Centered SI page, updated 2 October 2026, still includes Belmont Principles in AI Research inside an ongoing program. A 2024 NIST news story says NIST researchers argued that Belmont-style human-subjects principles could apply…
The NIST record here does not show a binding rule that AI research must follow human subjects ethics rules. Instead, NIST’s Human-Centered SI program page, updated 2 October 2026, still lists Belmont Principles in AI Research inside an ongoing program, while NIST’s 15 February 2024 news story says a NIST-authored paper argued that Belmont’s principles could be applied to AI research involving human subjects. The same NIST story says the paper was not official guidance and that the regulations requiring Belmont apply only to government research.
What NIST’s Human-Centered SI page says
NIST’s Human-Centered SI page describes an ongoing program and says: “The Human-Centered Super Intelligence (SI) program within the NIST Human-Centered Technologies Group performs research across several SI areas, always keeping humans at the core of our work.” The same page lists several projects under that program, including Belmont Principles in AI Research, alongside work on evaluation, risk assessment, trust measurement and taxonomy.
For the Belmont item, the NIST page says the team collaborated with NIST’s Research Protections Office to investigate how the Belmont Principles could apply to AI research. The page defines Belmont’s three principles as respect for persons, beneficence and justice, and it says the team also conducted qualitative research with the Research Protections Office on decisions to participate in human-subjects research involving AI.
NIST’s page adds that the team conducted semi-structured interviews with federal employees to identify factors in consent for a human-AI interaction study, and it says results are forthcoming. The same entry also states: “This work was completed prior to the issuance of the September 29, 2026, Executive Order 14434: Inaugurating the Era of Super Intelligence.”
One reading is that NIST is still keeping the Belmont frame visible inside a current federal research program even though the specific work described there predates the 29 September 2026 executive order. Because the page remains live, updated and grouped with other human-centered measurement work, it keeps human-subjects-style ethics in the same conversation as federal AI evaluation and risk methods.
What the 2024 NIST article argued
NIST’s 15 February 2024 news story says a team of NIST researchers published a paper in Computer arguing that the Belmont Report could provide a basis for ethical AI research. The story adds that the paper was the authors’ own work rather than official NIST guidance, even as it fit with NIST’s broader work on trustworthy and responsible AI.
Kristen Greene, a NIST social scientist and co-author, said: “We looked at existing principles of human subjects research and explored how they could apply to AI.” Greene added: “There’s no need to reinvent the wheel. We can apply an established paradigm to make sure we are being transparent with research participants, as their data may be used to train AI.” In the same story, the NIST research team states: “Applying the Belmont Report’s principles to human subjects in AI research could bring us closer to trustworthy and responsible use of AI.”
The article also explains the paper’s limits. The NIST research team states: “The regulations that require application of the Belmont Report’s principles apply only to government research.” The story says the authors nevertheless suggested broader use of those concepts in research involving human subjects, including cases where training data may come from people who did not consent to that use.
NIST’s story gives two concrete ethics concerns. It says lack of consent for data use can raise a respect-for-persons issue. It also says inappropriate exclusion can bias datasets against some demographics, which the authors connect to Belmont’s justice principle.
The case for using human-subjects ethics rules in AI research
The strongest case in this record rests on continuity and fit. NIST’s program page keeps the Belmont project on a current federal page devoted to human-centered SI, and the 2024 NIST story says the researchers were trying to move from general AI principles to more established research practice.
That matters because Belmont does not just offer abstract values. NIST’s story says the framework can translate into informed consent, risk minimisation and fair subject selection in AI-related studies. Greene’s argument is that an older research-ethics model can be reused rather than rebuilt from scratch.
Another reason this signal matters is the kind of AI work NIST describes. The program page says the Belmont work included interviews about consent in a human-AI interaction study. The 2024 story also points to people whose data may be used to train AI. In that sense, the question is not only about model performance. It overlaps with wider disputes about whether AI can consent, AI welfare versus AI rights, and whether AI should have rights.
One careful reading, then, is that NIST is treating AI research ethics partly as a research-governance question, not only as a product-safety question. The page’s placement of Belmont beside evaluation and risk-assessment projects supports that reading, even though the page does not say that AI systems themselves are human subjects or rights-holders.
The case for reading this as a limited signal
The counter-reading is also strong on this record. NIST’s 2024 story expressly says the paper was not official guidance. Greene also says: “We’re not advocating more government regulation. We’re advocating thoughtfulness.” That cuts against reading the article or the project page as a new federal mandate.
The same story further narrows the claim by saying the existing regulations that require Belmont apply only to government research. Greene adds elsewhere in the article that private-sector adoption is voluntary. So the material here supports an argument for using Belmont-style review more widely, but not a claim that NIST has imposed that review on industry.
The program page also stops short of describing a fresh policy step. It says the Belmont work was completed before the 29 September 2026 executive order, and it says results from the consent interviews are forthcoming. On that reading, what remains current is the project’s presence on the page and its inclusion in an ongoing program, not a newly issued rule, standard or enforcement position.
So the narrow answer to the headline question is that NIST is keeping the human-subjects ethics frame alive in federal AI standards discussion, but only as a research line and a cited precedent. The stronger claim that AI research now must follow those rules is not supported by the NIST documents here.
Frequently asked questions
Has NIST made Belmont-style review a binding rule for all AI research?
No. NIST’s 15 February 2024 news story says the paper was the authors’ own work, not official NIST guidance, and the same story says the regulations requiring Belmont principles apply only to government research.
What does NIST say the Belmont Principles are?
NIST’s Human-Centered SI page says the Belmont Principles are respect for persons, beneficence and justice, and the page describes them as principles for protecting human subjects in federally funded research.
Is the Belmont project still active at NIST?
NIST’s Human-Centered SI page, updated 2 October 2026, still lists Belmont Principles in AI Research inside an ongoing program. But the same project entry says the work was completed before the 29 September 2026 executive order.
Why would AI research use human-subjects ethics at all?
NIST’s 2024 story says the point is to address issues such as informed consent, minimising risk, and avoiding inappropriate exclusion in datasets. The story also says people’s data may be used to train AI even when they did not consent to that use.
Caveats: The claims here about the Human-Centered SI page’s current project list, updated date and ongoing program status come from that page alone. The characterization of the 2024 paper as not official NIST guidance, and the explanation that Belmont’s binding regulations apply only to government research, come from NIST’s 15 February 2024 news story.
Sources
- NIST — Human-Centered SI — 2026-10-02
- NIST — NIST Researchers Suggest Historical Precedent for Ethical AI Research — 2024-02-15