No federal standard for assisted living — at least USD 12 billion of federal money reached the sector in 2024

What is happening?

The United States has no federal quality or safety standard for assisted living facilities, and no federal definition of what one is. In calendar year 2024 federal Medicaid and Medicare paid at least USD 12 billion for services delivered in these settings — a figure the Government Accountability Office published on 2026-06-02 and released publicly on 2026-07-02, while stating in the same report that it is an undercount. Assisted living is licensed and regulated at the state level, unlike nursing homes, which sit under an extensive federal regime.

The 2024 money divides into USD 3.5 billion of federal Medicaid spending, part of a combined federal and state total of USD 6.2 billion, and USD 8.5 billion of traditional Medicare spending — hospice 6.1, home health 1.4, everything else 0.9. About 300,000 Medicaid beneficiaries and about 829,000 Medicare beneficiaries received services in assisted living that year. As of March 2025, 44 states covered assisted living services under Medicaid, 29 of them through home and community based services waivers.

About 1,000,000 Americans lived in assisted living facilities in 2022, the figure GAO cites from the federal National Post-acute and Long-term Care Study. The absence of a federal standard is not a new observation about them. GAO examined four states in 1999, examined Medicaid oversight in 2018, and in 2026 produced a report that carries no recommendations at all — closing instead with the position of CMS that any change to federally funded assisted living services would likely require congressional or individual state action.

Whose problem is this?

The people who would benefit from a standard are dispersed across 51 regulatory regimes and pay privately, so there is no fiscal constituency demanding value for federal money. The organizations that would bear a standard are national, funded, and organized.

Where does this problem end?

The boundary matters because the money and the hook are structurally decoupled. Federal dollars reach the setting in quantity every year without creating a single federal relationship with a single facility, so the ordinary route by which the federal government regulates a care setting is not narrow here. It is absent.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a target, a threshold or a date. Not one document says what level of federal spending, what number of deaths, or what year would warrant a federal standard. The senators describe little federal oversight against billions in federal funding; the trade association describes the same USD 12 billion as roughly 0.2 percent of federal Medicaid spending and well below one percent of the combined Medicare and Medicaid total. Both readings are arithmetically true of the same number, and no source proposes a rule for which reading governs.

How big is it?

Between 818,800 and 1,200,000 people, with the federal estimate near 1,000,000. The central figure is the one GAO cites from the federal NPALS survey for 2022. The two ends of the interval are different kinds of quantity, and that is itself the finding.

The upper end, 1,200,000, is a count of licensed beds rather than of people, reported in a January 2024 release. Capacity is an upper bound on point-in-time occupancy, so that end overstates the number of residents by whatever the vacancy rate is, and no source opened here publishes the vacancy rate. The lower end, 818,800, is a resident count from the same January 2024 release, which states no data year for it. That number sits below the 2022 federal estimate, which is the signal that it is an earlier survey wave resurfacing in testimony rather than a 2024 measurement. It was being used in the venue where the policy question was being argued, and the record does not show anyone asking what year it described.

There is no federal list of these buildings. The gap between 1,000,000 and 1,200,000 is definitional, the gap between 818,800 and 1,000,000 is chronological and undated, and nothing exists to adjudicate either one. For a sector into which the federal government paid at least USD 12 billion in a single year, the count of people living in it is uncertain by roughly 380,000.

The federal payment relationship touches a subset: about 300,000 Medicaid beneficiaries and about 829,000 Medicare beneficiaries in 2024, heavily overlapping because most of the Medicaid group is also Medicare eligible. But the absent thing applies to the facility, so exposure is universal within the setting regardless of who pays.

Under what conditions does it arise?

1. Federal authority over a care setting runs through the payment relationship, and here there is none. CMS regulates nursing homes through conditions of participation — the government buys the care and can stop buying. Medicaid is barred by statute from paying room and board in assisted living, and pays only for services, through state-administered waiver and state plan authorities where the state sets eligibility, defines the service, qualifies the provider, sets the rate and monitors quality. Medicare pays separately certified third parties, hospices and home health agencies and physicians, who would be equally payable if the same person lived in a house.

2. The household is the payer of record, which removes the ordinary political route as well. Most residents and their families pay privately at a national median of USD 6,200 a month in the 2025 survey. There is no constituency demanding value for federal money, because federal money is not what is mainly at stake for the people affected.

3. The absence of a standard produces the absence of the evidence that would argue for one. No federal standard means no federal reporting requirement, which means no national database, which means the only national count of a specific harm in this sector is a newspaper investigation that describes itself as incomplete for about 40 percent of the resident population. The central ask of the leading advocate at the 2024 hearing was not a standard but a database, because the case for a standard requires evidence that only a standard would generate.

4. The exemption is defended in writing, by more than one organization, and competently. A trade association published a full doctrinal defense on 2026-07-10, eight days after the GAO release and naming the report as the document that may be cited to justify federalization. A second association with different members states that it actively opposes expanded federal requirements. The defense costs money and staff, which is evidence that something is at stake.

5. The venue where agreement is available absorbs the attention. Affordability and workforce shortage are real adjacent problems that both sides work on, and every 2025 and 2026 federal bill found in this round moves money rather than setting standards. The question that has no available consensus is the one that does not get asked.

What has been tried?

Two directions have been tried across twenty-seven years — set a federal standard, and let the states do it. The first failed once, at the definition, and has not been attempted since. The second is still running, and the most complete national compilation of its output states that it did not standardize terminology across states and that a blank field must not be read as the absence of a requirement.

What was found?

Why is it still unsolved?

Institutional exemption — the class is carved out of a federal regime that exists, by the design of the statutes that route federal money to it.

The federal government never buys the thing it would need to regulate. Its leverage over nursing homes is conditions of participation, and a single certified bed pulls an entire building under federal survey protocols and more than 150 violation codes, as the industry itself describes the arrangement. In assisted living none of that machinery can attach. Medicaid may not pay room and board, so the residence is never purchased; Medicaid pays only for services, and pays them through authorities the state administers end to end. Medicare pays hospices and home health agencies and physicians who are already federally certified in their own right and who would be paid identically if the beneficiary lived in an apartment. So USD 12 billion can move into the sector in one year without producing one federal counterparty relationship with one assisted living facility. A lawyer for an advocacy affiliate put the consequence in a single clause to reporters in December 2023 — without the ability to withhold payment, there is no hook.

The one time Congress substituted consensus for authority, consensus broke at exactly the point where authority would have been required. The Assisted Living Workgroup ran from fall 2001 to spring 2003 with nearly 50 member organizations and adopted 110 of 127 recommendations by supermajority. What it could not adopt was a definition of assisted living, along with private room requirements, defined levels of care and dementia training hours. A voluntary body cannot adopt a definition, because the definition is the boundary of the regulated class and every member sits on one side of it. The chairman told the hearing of 2003-04-29 that the report did not present a uniform set of model rules. The output went to the states as a reference guide, and that is the moment the exemption stopped being an accident and became a decision.

The pattern is stable because the absence reproduces the evidence vacuum that then defeats the case for closing it, and because 2026 is writing the deference into statute. With no standard there is no reporting requirement, with no reporting requirement there is no national repository, and the only national count of harm in this sector was assembled by reporters who scraped more than 160,000 state inspection reports and still came up incomplete for about 40 percent of the population, in a sector where ten states neither provide nor publish such records. Meanwhile the live federal vehicle, introduced 2026-04-30 in the Senate and 2026-05-04 in the House, would make assisted living a mandatory Medicaid benefit while defining the covered service as whatever is consistent with State law. If it passed as introduced, the federal government would become a far larger payer and would write its own non-jurisdiction into the Social Security Act in the same sentence.

What observation would mean it is solved?

Candidates — (a) Congress enacts a federal definition and a federal quality, staffing and reporting standard for assisted living, with an effective date (b) a federal count of residents, facilities and critical incidents is published annually and is complete for every state (c) the harm measures that can be counted, such as unattended departures and deaths in these settings, fall for several consecutive years.

(a) alone is weaker than it looks, and the weakness has a name. Nothing in either 2026 bill creates a survey agency, a certification process, or an appropriation to run one, and no source opened here raises the question of by what apparatus a federal standard would be enforced — not even the people asking for the standard. A standard with no apparatus is paper. The adjacent sector shows what happens next: the annual report of the same association for 2025, a document that is not among the sources opened here and is cited only as it was described in this round, lists the defeat of a federal nursing home staffing mandate among the accomplishments of that year — in a sector where federal jurisdiction is uncontested.

(b) alone counts data, not care. A complete national dataset would end the loop described above, which is why the advocates asked for it first, but measurement is not protection. The 2018 finding that 26 of 48 state Medicaid agencies could not produce a critical incident count has never been retired, and a national dataset on state violations already exists in private hands, purchased by trade associations from a commercial vendor and published by nobody.

(c) alone cannot be read on its own. The only national harm count in existence describes itself as an undercount, and better counting would push the measured number up while the underlying harm fell. Any fall would also have to be read against the age wave. The population aged 80 and over stood at 14.7 million in 2025 and is projected to be roughly 28 percent larger by 2030, so a flat count would represent a falling rate. The three have to be read together, and (b) has to come first for (c) to mean anything.

What is it connected to?

Fills with researchlong-term care financing and the absence of a public insurance route for it, nursing home regulation and the contested federal staffing standard, the direct care workforce and its wage structure, housing policy for older adults including the low income housing tax credit, and the same regulatory boundary in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How many assisted living facilities exist. GAO in 2026, with access to CMS, HUD, Social Security and Veterans Affairs data, produced no national facility count. The figures in circulation, roughly 30,000 and 30,600, trace to industry compilations and a sample survey rather than to a register. Every other number in this dossier rests on a denominator that nobody owns.
• How many people are harmed or die in these settings in a year. The December 2023 investigation states plainly that the exact number cannot be known because nobody is counting. No federal agency publishes such a figure, no state is required to report one federally, and the 2018 finding about state incident tracking still stands.
• The size of the undercount inside the USD 12 billion. GAO lists the reasons the figure is low — Medicare Advantage excluded, though about half of Medicare beneficiaries were enrolled in it in 2024; eight named states where only a portion of spending was captured; housing, veterans and supplemental income amounts described as unknown — and then publishes a single point with no range and no ceiling. Every argument about proportionality is being conducted against the floor of an interval whose top nobody has estimated.
• What became of the 110 recommendations adopted in 2003. They went to the states in April 2003. Twenty-three years later no source read here counts how many states adopted how many of them, including the 51-jurisdiction industry review, which tracks changes made during 2025 but benchmarks no state against the 2003 work. The one federal attempt was handed to a venue that has never reported back.
• Whether anybody has priced any of this. The trade association asserts that federal standards would raise costs, shrink Medicaid participation and worsen supply, with no estimate and no model attached. Advocates asserting the benefit publish no figure either. The ACCESS Act would create a mandatory Medicaid entitlement and carries no cost estimate in any source found here.
• Who would enforce a federal standard. No source raises the question of what body would survey, certify or sanction, or where the money for it would come from.
• Why the national violation dataset is private. Two trade associations buy state safety and violation data from a commercial vendor, publish none of it, and declined to provide it to reporters. The data is compiled from public state records, and no source discusses whether the public has any claim on it.
• How the state regimes compare. The most complete national compilation states that it did not standardize terminology across states and warns that a missing entry must not be read as a missing requirement. So the one document that would let anyone compare 51 regimes is explicitly not a comparison, and no comparable 51-jurisdiction table of staffing, training or inspection requirements exists from any source.

See the evidence

More primary material was read here than the evidence tier suggests. The full text of GAO-26-107884 supplies nearly every money and scale figure; the introduced texts of S. 4479, H.R. 8662 and H.R. 3000 were read directly and are the basis for the finding that the live 2026 vehicles create no federal standards; and the 2003 hearing record is the primary account of where the one federal attempt broke. The tier stays at secondary because two load-bearing primary sources could not be reached at all — the hearing record of 2024-01-25 returned 403 on every path, so what that forum produced is taken from the Gerontological Society of America release and from KFF Health News, and the NPALS survey pages returned 403, so the figure of about 1,000,000 residents is used as GAO reports it and the 818,800 figure could not be traced to a survey wave. The December 2023 investigation was read as a verbatim reprint hosted by an industry association rather than at the publisher, and the delay dates for the Medicaid access rule come from the tracking of a trade association because the agency pages returned 403. Where sources disagree the disagreement is left visible rather than resolved. One outlet covering the GAO release gives USD 8.7 billion of Medicare spending where GAO gives 8.5, and 40 waiver states where GAO gives 29; the GAO report is treated as controlling and the drift is recorded because it appeared within five days of publication. The national median cost appears as about USD 6,000 a month in December 2023, USD 54,000 a year in January 2024 and USD 6,200 a month in the 2025 survey, and no source states which figure a household should plan against. Most fundamentally, the same USD 12 billion is read by the requesting senators as billions with little oversight and by the trade association as 0.2 percent of federal Medicaid, and no source found offers a rule for which reading governs. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 818,800–1,200,000 As of 2022 to 2024, data years not uniformly stated

Sensitivity The upper bound is a bed count, not a person count. Licensed capacity exceeds point-in-time occupancy by an unknown margin, so 1,200,000 overstates the number of residents and no source opened here publishes the vacancy rate that would close the conversion. The width of roughly 380,000 is made of two different kinds of uncertainty. Between 1,000,000 and 1,200,000 the spread is definitional, residents against licensed beds. Between 818,800 and 1,000,000 the spread is chronological and undated, a figure whose data year no source states being quoted as current in a Senate hearing two years after a larger federal estimate. Nothing exists to adjudicate either gap, because there is no federal register of these facilities and GAO in 2026 could produce no national facility count. The interval also fails to count several groups. It is point-in-time, so it misses annual throughput, and no source found publishes an admissions or turnover figure. It excludes the households that are the payer of record at a national median of USD 6,200 a month in the 2025 cost survey. It excludes the workforce, reported at 478,500 people with two thirds in direct care in a January 2024 release that states no data year, who work under training rules set by 51 separate regimes with no federal floor. It excludes settings that states license under other names such as residential care, personal care home and board and care, whose inclusion depends on which state licensed the building and which compilation was consulted, and it excludes settings outside state licensure entirely. In the opposite direction the interval cannot be corrected downward for people already protected by a strong state rule, because no comparable table of state requirements exists. Finally, the population aged 80 and over stood at 14.7 million in 2025 and is projected to be about 28 percent larger by 2030, so the exposed count is scheduled to grow whether or not anything else changes.

Regional breakdown No state-level resident count could be confirmed. There is no federal register or list of assisted living facilities, and the most complete national compilation of state rules states that it did not standardize assisted living terminology across states, so the state entries are not comparable to one another, as that compilation itself states. Splitting the national figure by state population would be proportional allocation and would be wrong, because these facilities are distributed with the population aged 80 and over, with state licensure definitions and with local housing supply rather than with population.