The US effort to set rules for testing artificial intelligence is running into a basic constraint: the agency asked to lead the work may not have enough money or time to do it on its own.
The National Institute of Standards and Technology (NIST) is central to the White House executive order on AI announced in October. That order calls for new standards to stress-test AI systems for bias, hidden threats, and rogue tendencies. But people familiar with the work say NIST does not have the budget needed to complete that task independently by the July 26, 2024, deadline.
A Big AI Mandate With A Tight Clock
NIST is not new to AI policy. In January 2023, it released an AI risk management framework meant to guide both business and government. The agency has also worked on ways to measure public trust in new AI tools.
The new White House assignment is broader and faster. NIST is expected to help define how AI models should be evaluated, how “red-teaming” (adversarially testing) models should work, and how US-allied nations might align around NIST standards. It is also asked to support plans for “advancing responsible global technical standards for AI development.”
At the NeurIPS AI conference in New Orleans last week, Elham Tabassi, associate director for emerging technologies at NIST, described the timeline as “an almost impossible deadline” for the agency.
That phrase captures the central tension. The US government wants reliable AI safety testing soon, but the technical field is still developing. According to a bipartisan open letter signed by six members of Congress on December 16, there is “significant disagreement” among AI experts over how to work on, measure, or define safety issues in AI systems.
Why Funding Shapes Independence
NIST standardizes a wide range of things, including food ingredients, radioactive materials, and atomic clocks. But when it comes to frontier AI, it is operating in a landscape where private companies have much larger resources tied directly to the technology.
OpenAI, Google, and Meta each likely spent upwards of $100 million to train the powerful language models behind applications such as ChatGPT, Bard, and Llama 2. NIST’s budget for 2023 was $1.6 billion, and the White House has requested that it be increased by 29 percent in 2024 for initiatives not directly related to AI.
Several sources familiar with the situation at NIST say the current budget will not stretch far enough for the agency to figure out AI safety testing on its own. That is where lawmakers’ concerns become sharper: if NIST cannot do the work independently, it may need to rely heavily on outside organizations, including companies that are developing the same AI systems the standards could affect.
The congressional letter said, “We have learned that NIST intends to make grants or awards to outside organizations for extramural research.” The lawmakers warned that there did not appear to be publicly available information about how those awards would be decided.
Transparency Is Becoming The Core Issue
The concern is not simply that outside experts may contribute. AI safety testing is complex, and NIST has already engaged with outside groups in earlier work. Discussions around its risk management framework, before the executive order, involved Microsoft; Anthropic, a startup formed by ex-OpenAI employees that is building cutting-edge AI models; Partnership on AI, which represents big tech companies; and the Future of Life Institute, a nonprofit dedicated to existential risk, among others.
The issue is whether the process for shaping standards will be visible enough, and whether the balance of influence will be clear. If companies with major AI projects help define testing methods, policymakers want to know how conflicts or competing interests will be handled.
NIST has taken at least one step toward a more open process. On December 19, it issued a request for information seeking input from outside experts and companies on standards for evaluating and red-teaming AI models. It is unclear whether that request was a response to the congressional letter.
NIST spokesperson Jennifer Huergo confirmed that the agency had received the letter and said it “will respond through the appropriate channels.”
Experts See Measurement As The Battlefield
Several AI experts quoted in the source article frame NIST as important precisely because it is a scientific body rather than a company selling AI products.
Rumman Chowdhury, a data scientist and CEO of Parity Consulting who specializes in testing AI models for bias and other problems, said, “As a nonpartisan scientific body, NIST is the best hope to cut through the hype and speculation around AI risk.” But she added, “But in order to do their job well, they need more than mandates and well wishes.”
Yacine Jernite, machine learning and society lead at Hugging Face, said NIST has far fewer resources than the companies building the most visible systems. He said, “NIST has done amazing work on helping manage the risks of AI, but the pressure to come up with immediate solutions for long-term problems makes their mission extremely difficult.” He also said, “They have significantly fewer resources than the companies developing the most visible AI systems.”
Margaret Mitchell, chief ethics scientist at Hugging Face, pointed to another obstacle: commercial AI models are becoming harder to inspect. She said the growing secrecy around those models makes measurement more difficult for an organization like NIST, adding, “We can’t improve what we can’t measure.”
What The Fight Reveals About AI Governance
The NIST debate shows that AI governance depends on more than executive orders and policy goals. Standards require technical work, expert judgment, public trust, and resources. If any one of those is weak, the resulting system may struggle to carry authority.
The White House executive order also calls for a new Artificial Intelligence Safety Institute to support the development of safe AI. In April, a UK taskforce focused on AI safety was announced with $126 million in seed funding.
For the US, the immediate question is whether NIST can define meaningful AI safety standards under an aggressive timeline without becoming too dependent on the companies whose systems may be evaluated. The answer will shape not only how AI models are tested, but who gets to decide what safe AI means in practice.