DOJ Reviewed Nine NSF STEM Education Programs and Killed Three. The $104 Million Is the Small Number — the Reallocation Rule Is the Big One.

On August 12, 2026, the Justice Department's Office of Legal Counsel delivered an opinion to the National Science Foundation reviewing nine of the agency's STEM education programs against the equal protection guarantee. It became public two days later.

Three programs did not survive. Five more may continue only if run without the eligibility criteria that defined them. One survived intact.

The dollar figure attached to the terminated programs is $104 million out of NSF's roughly $938 million STEM education portfolio — about 11 percent. That number will anchor most of the coverage, and it understates the opinion considerably, because the OLC did not stop at the programs. It reached the money.

The three programs held unconstitutional

Louis Stokes Alliances for Minority Participation (LSAMP). Established in 1990 and named for the Ohio congressman who championed it, LSAMP funds consortia of colleges — "alliances" — that build pipelines into STEM bachelor's degrees for students from groups historically underrepresented in those fields. It is, by a wide margin, the longest-running program on the list and the one with the deepest institutional infrastructure. Some alliances have been continuously funded for three decades.

Alliances for Graduate Education and the Professoriate (AGEP). The graduate-and-faculty analogue: institutional policy change, mentoring structures, and faculty development aimed at diversifying the STEM professoriate.

The Hispanic-Serving Institution track of Improving Undergraduate STEM Education (IUSE). Grants available exclusively to HSIs to build STEM capacity and raise graduation rates.

The OLC concluded all three violate the equal protection guarantee "in their entirety." Not in application, not in a subset of awards — as designed.

The legal theory

The opinion's core move is a narrow reading of when the federal government may take race-conscious action at all. Under OLC's framing, an agency may use a racial classification only to remedy specific, identified past discrimination by that actor — not to address general societal disparities, however well documented. The 2023 Supreme Court decision in Students for Fair Admissions v. Harvard is the cited authority.

Applied to LSAMP and AGEP, the theory is close to dispositive by construction. These programs were built on a diagnosis of aggregate underrepresentation across a field, not a finding that NSF itself discriminated against anyone. There is no version of them that satisfies a remedial-only standard.

Critics have pressed on exactly that construction. Ebony O. McGee, a Bloomberg Distinguished Professor at Johns Hopkins, argued that the standard is engineered to be unmeetable — that discrimination in STEM "hides behind the values of objectivity, individualism, hypercompetitiveness," and does not generate the kind of discrete, attributable findings the test demands. That is a real critique of the doctrine. It is not, at the moment, a defense available to a grantee.

The five programs on conditions

Five programs may continue, stripped of the criteria at issue:

• ADVANCE — permitted if sex-based preferences are removed. ADVANCE exists to address the underrepresentation of women in academic science and engineering careers, so "remove the sex-based preferences" is a substantial redesign, not a paperwork edit.
• Advanced Technological Education (ATE) — permitted without race- or sex-based preferences. ATE, which funds two-year college technician education, is the least affected of the five; its core mission is not defined by a protected characteristic.
• Graduate Research Fellowship Program (GRFP), Advancing Informal STEM Learning (AISL), and the Robert Noyce Teacher Scholarship Program — must be administered without consideration of race or sex.

For GRFP in particular, this codifies a direction NSF had already been moving. It is a reminder that the country's most prominent graduate fellowship now operates under an explicit federal legal opinion governing what its reviewers may weigh.

The Tribal Colleges and Universities Program survived the review. The opinion does not appear to disturb it, and the likely reason is doctrinal rather than political: classifications tied to federally recognized tribes have long been treated as political rather than racial, on the Morton v. Mancari line. Institutions should not read TCUP's survival as a signal that the reasoning is soft anywhere else.

The part that is not about programs

Here is the holding that will outlast the news cycle.

OLC concluded that the congressional suballocations directing funds to these programs are themselves unconstitutional — and that NSF may therefore treat those amounts as residual funds, available for other lawful STEM activities.

Read that carefully. It is not a conclusion that NSF must return the money, or that the money is impounded, or that it awaits a fresh appropriation. It is a conclusion that a line Congress wrote to direct roughly $104 million to specific purposes has no constitutional force, and that the agency now has discretion over where it goes.

The practical import for institutions: the money did not leave NSF. It is inside the same STEM education directorate, unencumbered. Whoever is positioned to compete for it in the next solicitation cycle will get it. That is a very different planning problem from a rescission, and a much shorter one.

The constitutional import is larger and will be litigated somewhere. An executive branch legal opinion that a congressional funding direction is void, and that the agency may consequently redirect the funds, is a claim about the separation of powers dressed as a claim about equal protection. It sits adjacent to every ongoing fight over impoundment and executive control of appropriated funds.

What institutions holding these awards should do now

Get the status of your specific award in writing. An OLC opinion is binding legal advice to the executive branch; it is not a termination notice. Existing LSAMP and AGEP awards are cooperative agreements and grants with terms, periods of performance, and — under current 2 CFR 200 — enumerated termination grounds. Ask your program officer, in writing, whether NSF intends to terminate, decline continuation, or allow the current budget period to run. Those three answers produce three different budgets.

Distinguish termination from non-continuation immediately. Most LSAMP and AGEP funding flows as annual continuations within a multi-year award. Declining to fund the next budget period is administratively far easier for an agency than terminating an active one, and it generates far weaker legal claims for you. If your award has a continuation decision due in the next two quarters, that is your live date — not any date in the opinion.

Protect the people first. These programs fund students and postdocs, not just institutional activity. Cost-sharing commitments, bridge funding, and institutional reserves are the tools that keep a cohort from being stranded mid-degree. Decide the order in which you will spend them before you need to.

Redesign toward criteria the opinion does not reach. The programs that survived on conditions are the template. Eligibility built on institutional characteristics that are not protected classifications — two-year colleges, rural-serving institutions, low-endowment institutions, first-generation-serving institutions, Pell-share thresholds — is untouched by this reasoning. So is eligibility built on individual economic circumstance. If your alliance's actual value is a functioning multi-campus transfer-and-mentoring pipeline, that pipeline can be described and funded on criteria that survive.

Compete for the residual funds. This is the unglamorous, correct move. Roughly $104 million is now discretionary inside NSF's education directorate, and the institutions that will capture it are the ones with a ready proposal when the next solicitation posts. Every institution that just lost an LSAMP alliance has, sitting in a drawer, a fully developed and independently evaluated model for moving students through undergraduate STEM. Rewriting that model against race-neutral eligibility is a weeks-long project, not a years-long one.

Diversify the funder base for this work. Private philanthropy operates under different constraints than a federal agency. Foundations with long-standing STEM-access portfolios are the obvious first calls, and they are, predictably, being called by everyone right now. Being early matters more than being persuasive.

The through-line

For thirty-six years, LSAMP was one of the most stable line items in federal STEM education. Its removal took a memorandum.

That is the operating condition now, and it applies well beyond the nine programs OLC reviewed. Any program whose eligibility criteria reference a protected characteristic is reviewable on the same theory, at any agency, on no notice. Institutions that build their next five years of programming on the assumption that a durable federal program is a durable funding source are making a bet the last eighteen months have repeatedly settled.

Design for the mission. Fund it from more than one place.