ACA Enrollment Fell to 19.2M in 2026. HHS Says It Was a Fraud Purge, Not Attrition

19.2 million people carried effectuated ACA Exchange coverage as of February 2026, down from 22.1 million at the close of 2025, and the Department of Health and Human Services says every one of those 2.9 million lost enrollees was improper or phantom, not price-driven attrition (HHS ASPE, June 2026). Improper enrollment itself fell from 5.6 million in 2025 to 2.6 million in 2026. If the composition claim holds, the residual pool got smaller and sicker in a very different way than a subsidy-driven exodus would produce.

Key Takeaways

• 19.2 million effectuated enrollees in February 2026, down from 22.1 million at the close of 2025 and still 85% above the February 2019 baseline of roughly 9.8 million.
• 2.9 million removed, split by ASPE into about 1.5 million blocked from subsidies they did not qualify for and 1.4 million ended through the elimination of year-round special enrollment pathways.
• Zero utilization, not low utilization, is what separates this cleanup from ordinary adverse selection. Paragon places roughly 6.4 million people in the fully subsidized 100-150% FPL band despite not having that income.
• 73% of returning 2026 enrollees worried about affording emergency care, and 44% said higher premiums made groceries, utilities or rent harder. Part of the exit was priced out, not cleaned out.
• 27.8% of zero-premium 2026 enrollees had coverage cancelled by May, evidence that a share of the 2026 base period was unstable membership before the year ended.

What the ASPE Brief Actually Measures

The ASPE issue brief of June 26 is doing something more specific than reporting a headcount. It is making a claim about who left, and that claim is the actuarially load-bearing part of the report. Its source is CMS's monthly effectuated enrollment series, cross-referenced against subsidy verification and program-integrity screening data.

ASPE traces improper and phantom enrollment growing from roughly 4.4 million in 2024 to a peak of 5.6 million in 2025, then falling to 2.6 million in 2026 as verification tightened, including more than 1 million enrollments tied to no valid Social Security number. It credits two program-integrity tracks for the 2.9 million removed: about 1.5 million enrollees blocked from subsidies they did not qualify for, and roughly 1.4 million ended or blocked through the elimination of year-round special enrollment pathways. The two tracks sum to the headline decline almost exactly, which is the numerical basis for the program-integrity framing.

The framing is contested by evidence the brief does not address. KFF's tracking survey found that among returning 2026 marketplace enrollees, 73% worried about affording emergency care or hospitalization and 44% said higher premiums made it harder to afford groceries, utilities or rent. At least some of the 13% decline was priced-out attrition. Two credible sources are reading the same 2.9 million through different lenses, and a 2027 filer has to weight them rather than choose.

Why the Distinction Is Zero Utilization, Not Low Utilization

The stronger evidence for ASPE's framing is not the enrollment count. It is claims data on what the removed population was costing. Paragon Health Institute's analysis of CMS enrollee-level claims files from 2019 through 2024 found the 100% to 150% of federal poverty level band, where net premiums run at or near zero, carried a dramatically elevated zero-utilization rate relative to every other income band. Paragon separately estimates roughly 6.4 million people were placed in that band after the 2025 open enrollment period despite not having that income.

Zero utilization rather than low utilization is the fact that does the work. A healthy 28-year-old who lapses because the premium tripled was a legitimate low-cost risk pricing favorably against the pool average, so removing them shifts the morbidity distribution upward. A phantom enrollment generated by broker duplication or income misreporting contributed almost nothing to the claims numerator, so removing it only ever diluted the denominator.

Single risk pool rate development projects a base-period claims experience forward with a trend and a morbidity adjustment, the mechanism the site's 2027 morbidity load explainer sets out. The adjustment answers one question: is the 2027 target population sicker or healthier than the 2026 base that generated the claims data? A headcount decline answers it badly, because the same decline is consistent with healthy attrition, proportional exit, or zero-cost removal.

The bounded arithmetic is what a filing has to state. Removing a genuinely zero-claims policy from a 22.1-million-member pool raises the survivors' average cost mechanically, in exact proportion to how far below the mean that member's true cost sat, and no further. A 2027 filing that assumes the entire 2.9 million was morbidity-neutral will understate residual claims cost; one that assumes it was all healthy lapse, on the Wakely-style model in the site's enrollment cliff analysis, will overstate it.

One datum narrows the weighting. Paragon's reading of the same underlying data found 27.8% of zero-premium 2026 enrollees had coverage cancelled by May, concentrated in the income band it flags for misreporting. A near-30% cancellation rate inside the zero-premium tier is consistent with a meaningful share of the 2026 base already being unstable, low-engagement membership. It shifts the actuarial prior toward the morbidity-neutral end; it does not resolve the dispute.

The Adverse Mechanism the Brief Does Not Describe

ASPE describes 2026. The mechanism that will dominate the 2027 filing cycle is a different one: the enhanced premium tax credits that held net premiums down through 2025 have expired, and the resulting affordability shock is a textbook adverse-selection driver rather than a program-integrity story. Subsidy-driven exit selects specifically against the healthiest members, the ones with the most price elasticity, which is the opposite selection profile from documentation screening.

Carrier-level morbidity adjustments in the site's subsidy-expiration filing coverage already run well above what a pure enrollment cleanup would justify. Conflating the two mechanisms, treating the whole 2.9 million as either benign cleanup or adverse selection, is the modeling error the brief's framing invites when read uncritically in either direction.

The two also present differently inside one book. A carrier whose 2026 losses concentrated among duplicate or unverifiable enrollments should see less morbidity deterioration in its 2027 base than one whose losses concentrated among price-sensitive subsidy-dependent members. Carrier-level effectuation and lapse data separates those populations for a given filer; the national aggregate does not, which is why the effectuation-based methodology Wakely published is built on the carrier-level distinction.

The cleaner roster helps in one narrow place and less than it appears. Risk adjustment transfers run on the same enrollee-level demographic and claims data program-integrity screening is scrubbing, so 2.6 million fewer improperly documented records means HHS-HCC scores computed from more reliable inputs, and somewhat more stable transfer estimates than a pure adverse-selection contraction would produce. But verification targets income and identity documentation, not diagnosis coding or claims completeness for the enrollees who remain. A cleaner enrollee roster is not a cleaner risk score, and the credibility gain sits in the eligibility layer of the data rather than the clinical layer.