Bipartisan NSF AI Education Act Sets Goal of Training One Million Workers by 2028 as Schools and Lawmakers Grapple with AI in the Classroom
Senators Cantwell and Moran introduce federal legislation to fund AI scholarships, create community college Centers of Excellence, and train one million workers, while a Stanford review finds only 20 rigorous studies on AI in K-12 and states introduce competing bills to regulate classroom use.
Overview
A bipartisan push to build an AI-literate workforce is gaining momentum in Congress, arriving alongside mounting evidence that neither K-12 schools nor universities have settled on how to integrate the technology they are racing to teach. In early March, Senators Maria Cantwell (D-Wash.) and Jerry Moran (R-Kan.) introduced the NSF AI Education Act of 2026, a bill that would channel National Science Foundation resources toward scholarships, community college AI centers, and a grand challenge to train one million American workers in AI by 2028. The legislation lands in a policy environment where a Stanford review has found that only 20 of more than 800 studies on AI in K-12 education meet the bar for high-quality causal research, and where at least 25 states have introduced bills this session to regulate or restrict AI use in classrooms.
What the Bill Would Do
The NSF AI Education Act targets the AI skills gap at multiple levels of the education pipeline. According to the Senate Commerce Committee announcement, the bill would authorize the NSF to award undergraduate and graduate scholarships in AI, with designated tracks for agriculture, education, and advanced manufacturing. It would also create professional development fellowships for workers already in the STEM and education workforce.
The bill’s most concrete structural provision calls for establishing at least five “Centers of AI Excellence” at community colleges and vocational schools across the country. These centers would develop and disseminate best practices for AI education, coordinated with the Regional Technology Hub Program. A summary released by Senator Moran’s office emphasized the bill’s focus on reaching underserved populations, including women, rural communities, and tribal students.
Separately, the bill authorizes the Secretary of Agriculture, in coordination with the NSF, to fund AI research and training through Land-Grant Universities and the Cooperative Extension Service, a network designed to reach rural communities that traditional tech workforce pipelines often miss.
The legislation’s headline target is an NSF Grand Challenge to educate one million or more workers in AI-related skills by 2028. The bill does not specify an appropriations figure, and the House companion, H.R. 5351, was introduced in an earlier session and referred to committee. The bill’s endorsers include OpenAI, Code.org, California State University Fresno, and several community college systems.
The Evidence Gap in K-12
The push for federal AI education investment arrives against a backdrop of limited evidence about what actually works in classrooms. A 2026 review by Stanford’s SCALE initiative examined more than 800 academic papers related to AI and K-12 education and found just 20 that meet the standard for rigorous causal research, meaning studies that can credibly establish whether an AI tool changed outcomes for students or educators.
Among those 20 studies, many find that students perform better on tasks like math practice, programming assignments, or writing when they can use AI tools during the activity. But evidence is mixed when students complete assessments without AI support: in some cases performance improves, in others it remains unchanged or declines. The review noted that AI tools designed with pedagogical guardrails, such as tutoring systems that provide hints or guide reasoning, show more promising outcomes than general-purpose chatbots that simply provide answers.
Critically, the Stanford team found no high-quality causal studies of student AI use conducted in U.S. K-12 classrooms. Most existing research examines short-term outcomes rather than long-term learning, and almost none examines impacts on equity, student wellbeing, or social development.
One study that has drawn attention for its rigor is a randomized controlled trial published in Scientific Reports from researchers at Harvard. The trial, conducted in a large undergraduate physics course with 194 students, found that students using a custom AI tutor informed by pedagogical best practices learned significantly more in less time compared to an active learning classroom setting, while also reporting higher engagement and motivation. The results suggest that thoughtfully designed AI tutoring can outperform traditional methods, but the study’s narrow scope, a single university course, underscores how far the field is from generalizable conclusions.
States Move to Regulate While Adoption Surges
While Congress pursues investment, state legislatures are moving to establish guardrails. At least 25 states have introduced a combined 52 bills addressing AI in classroom instruction during the current legislative session. The approaches vary widely.
New York’s A.9190 would prohibit most classroom AI use below ninth grade, with limited exceptions. South Carolina’s H.B. 5253 proposes some of the strongest statewide requirements, including written parental opt-in consent, annual public disclosure of AI tools and data practices, and a prohibition on AI replacing licensed teachers in core instruction or grading. The bill would also ban automated high-stakes decisions about students without meaningful human oversight.
Utah has taken a different tack with S.B. 322, which would create a five-year educational technology regulatory sandbox allowing voluntary pilot programs under strict privacy, safety, and human-in-the-loop requirements. Statewide expansion would depend on independent evaluation and legislative approval.
The regulatory activity reflects the pace of adoption. About 85 percent of undergraduates now use AI for coursework, according to a poll by Inside Higher Ed and the Generation Lab cited by NPR, and student use of AI for school-related purposes has jumped 26 percent since last school year. Faculty adoption has risen 21 percent over the same period, but institutions remain divided on policy. As NPR reported, professors range from treating AI-generated essays as the equivalent of “bringing a forklift to the gym” to embracing generative AI as a collaborative tool that enhances student learning.
International Parallel: The UK’s AI Tutoring Pilot
The United States is not alone in pursuing government-backed AI education initiatives. In January, the UK Department for Education announced plans to develop AI-powered tutoring tools targeting up to 450,000 disadvantaged pupils in years 9 through 11 who are eligible for free school meals. The program will begin co-creation with teachers in summer 2026, with trials in secondary schools later this year and tools available by the end of 2027.
The UK initiative is grounded in evidence that one-to-one tutoring can accelerate a pupil’s learning by approximately five months, and it aims to narrow an attainment gap in which only one in four disadvantaged students achieves a pass at grade 5 or above in English and maths at GCSE, compared to over half of their peers. The government has emphasized that the tools are intended to complement face-to-face teaching, not replace it.
What We Do Not Know
The NSF AI Education Act does not include a specific appropriations figure, and the bill’s path through a Congress that has struggled to pass comprehensive AI legislation remains uncertain. The one-million-worker training target by 2028 is ambitious, but the bill does not define what level of AI competency would count toward that goal.
Stanford’s finding that only 20 rigorous causal studies exist across the entire field of AI in K-12 education means that both federal investment and state regulation are proceeding largely in advance of the evidence. The Harvard tutoring study, while promising, examined a single physics course at an elite university; whether its findings transfer to the K-12 settings, underserved populations, and diverse subject areas that the NSF bill targets is an open question.
The gap between adoption velocity and evidence availability may be the defining tension of AI in education in 2026. Students and teachers are already using the tools at scale. The question now is whether policy can establish a framework, built on evidence that does not yet fully exist, before the window for shaping adoption closes.