Compliance reference · 29 USC § 794 · 34 CFR Part 104

Section 504 captions: federally funded programmes and the discrimination test for inaccessible video

Section 504 of the Rehabilitation Act is the older, broader sibling of Section 508. Where 508 sets a procurement standard for technology the federal government buys, 504 sets a civil-rights baseline for every programme that receives federal financial assistance — universities, hospitals, state and county agencies, federal-grant-funded research centres, primary and secondary schools that take Title I money. The question 504 asks about training video is not "did you tick the captions box?" It is "does a person with a hearing disability have meaningfully equal access to this content?" The captioning bar that flows from that question is functional, not formulaic — and that is exactly why the proper-noun failure mode of generic auto-captioning is the one that gets cited.

TL;DR

Section 504 (29 USC § 794) prohibits disability discrimination in any programme or activity receiving federal financial assistance. The implementing rules — 34 CFR Part 104 for the Department of Education, 45 CFR Part 84 for HHS, and parallel rules at every grant-making agency — bind universities, hospitals, school districts, state and county human-services agencies, and any organisation receiving federal grants or sub-grants. The legal standard is "equal opportunity to participate in and benefit from" the programme; for training video that means captions accurate enough that a deaf or hard-of-hearing learner gets the same training a hearing learner gets. The Department of Education's Office for Civil Rights (OCR) investigates 504 complaints; the typical finding when YouTube auto-captioning is the only caption track is that the captions exist but mangle the very terms the training is teaching, which fails the functional-access test. The fix is a captioning workflow that gets the proper nouns right on the first export.

Who is covered by Section 504

Section 504's reach is defined by the funding flow, not the entity type. The covered universe is wider than most operators expect:

Public and most private universities. Receive federal student aid (Title IV), federal research grants (NIH, NSF, DOE, DARPA), federal contract overhead, or work-study funding. Effectively every accredited US institution.
Hospitals and healthcare organisations. Receive Medicare and Medicaid reimbursement (operationally federal financial assistance under 45 CFR § 84), federal grant funding, or Veterans' Health Administration sub-contracts. Standalone medical training programmes that take federal money are independently in scope.
K-12 school districts and state education agencies. Receive Title I, IDEA Part B, and a long list of categorical grants. Their compliance training, professional development, and instructional video catalogues are all covered.
State and county human-services agencies. Receive HHS, USDA, HUD, and Labor block grants. The internal training programmes their case-workers complete fall under 504 the moment any federal dollar touches the programme.
Federal grant-funded non-profits. Workforce development organisations, mental-health programmes, food assistance non-profits — all in scope when they take federal grant money for the programme that produces or uses the video.
Federal contractors as flow-down recipients. Where the federal contract is structured as financial assistance rather than a procurement contract (this is fact-specific), 504 reaches the contractor's internal programme.

The reach goes further than 508. 508 is procurement-bound — it asks whether the federal government bought the technology. 504 is funding-bound — it asks whether federal money supports the programme. A university's internal LMS used to deliver training to its own employees is not 508-bound (the federal government did not buy it), but it is 504-bound the moment Title IV student aid money flows through the institution.

504 vs 508: the question each one asks

The two statutes are commonly conflated, and the conflation costs operators time. The cleanest framing:

Section 504 (1973) is a civil-rights statute. It asks: does the programme discriminate against people with disabilities? The remedy is whatever is needed to provide equal access, on a fact-specific basis. Captions for video are a primary example of an accommodation the rules treat as effectively required.
Section 508 (1986; revised 1998 and 2018) is a procurement standard. It asks: does the technology the federal government bought conform to the technical accessibility standards at 36 CFR Part 1194? The remedy is contract-level — the contracting officer can refuse delivery.

For a federally funded university running an internal employee-training programme on its LMS:

504 applies because the institution receives federal financial assistance. Captions are required as an equal-access matter. OCR is the enforcer.
508 may also apply to specific procurements (the LMS itself, a captioning vendor) if those purchases are tied to a federal contract or pass-through grant condition. The contracting office is the enforcer there.
If the institution is a state or local government entity, ADA Title II applies on top, with WCAG 2.1 AA as the explicit DOJ-rule standard since 2026-04-24 (large entities).

The captioning bar is consistent across regimes — synchronized, accurate, machine-readable. The audit mechanism differs.

The functional-access standard: why "captions exist" isn't the answer

The text of 34 CFR § 104.4(b)(1)(ii) is unusually clear: a recipient may not "afford a qualified handicapped person an opportunity to participate in or benefit from the aid, benefit, or service that is not equal to that afforded others." OCR, the courts interpreting the rule, and HHS's parallel guidance under 45 CFR § 84.4 read this as a functional standard: the question is what the user actually got, not whether the technical artefact existed.

For a training video, that turns into three sub-questions:

Were captions present at all? Bare minimum. YouTube auto-captions count for this question, however badly.
Were the captions synchronised within the SC 1.2.2 tolerance? Whisper auto-captioning generally is; rough machine output sometimes drifts.
Did the captions accurately convey what the speaker said, including the technical and proper-noun terms the training is teaching? This is where YouTube auto-captions typically fail and where OCR's findings most often land.

The third question is what the functional-access standard puts pressure on. A compliance training video for HIPAA workforce training that captions "PHI" as "fee" doesn't teach the trainee what they need to know — the discriminatory effect is the missed learning, not the missing file.

How OCR investigates 504 caption complaints

The Department of Education's Office for Civil Rights handles the largest share of 504 complaints, with parallel offices at HHS, DOJ Civil Rights Division, and other grant-making agencies handling sector-specific cases. The investigation pattern is consistent:

Complaint intake. Filed by a student, employee, contractor, or third-party advocate. The complainant often names specific videos and specific terms that were mangled. OCR opens an investigation under 34 CFR § 104.61 if the complaint states a 504 claim.
Document request. The recipient is asked for a representative sample of training-video URLs, the caption tracks attached, the captioning policy and workflow, and any prior complaints or remediation. Recipients with a documented workflow (vendor name, accuracy SLA, reviewer step) move through this stage faster.
Sampling. OCR investigators open the videos, run captions, and watch for whether a hearing-impaired learner could follow along. A 30-second sample with three mangled key terms typically produces a finding.
Resolution agreement. If a finding lands, the agency negotiates a resolution agreement: typically a remediation timeline, a captioning vendor commitment, a reviewer step in the workflow, and reporting back. OCR has the authority to refer to DOJ for funding-termination action under 34 CFR § 104.61, though this is rare; the operational consequence is usually reputational and the cost of the remediation work itself.

The pattern OCR investigators look for in the sampling step is the same pattern an ADA Title II complainant or a 508 audit catches: high-meaning words mangled, learning compromised. Generic auto-captioning fails this test on technical content with predictable regularity.

The proper-noun failure mode in 504-bound training content

504 reaches the broadest swath of L&D programmes — university HR onboarding, hospital clinical training, school-district employee compliance modules, state agency casework training. Each surface has its own proper-noun density:

University HR onboarding — federal regulatory citations (FERPA, Title IX, Clery Act, IDEA), grant-management acronyms (OMB Uniform Guidance, F&A rates), and institution-specific programme names ("Honors College", "Mead-Berkeley Initiative"). YouTube auto-captioning typically loses Roman numerals and hyphenated initiative names.
Clinical training in hospitals — drug names ("tirzepatide", "semaglutide", "evolocumab"), procedure codes (CPT, HCPCS, ICD-10), and provider acronyms (PCMH, ACO, HEDIS). The companion medical training captions reference walks the drug-name failure mode in detail.
School-district staff training — federal-programme acronyms (Title I, IDEA, ESSER, IEP), curricular framework names (NGSS, Common Core, B.E.S.T.), and assessment names that mangle in unique ways ("STAAR" → "stair", "MCAS" → "M cause").
HHS-grant non-profit training — agency programme codes (SAMHSA SOR, HRSA RWHAP, ACF TANF), state-agency acronyms (DCFS, DSHS, DHHS), and grant-vehicle names ("CCDBG", "TANF MOE").

The pattern repeats across surfaces: the words a 504-covered learner most needs to retain are exactly the words generic STT mangles.

The glossary-biased workflow for 504-covered programmes

Build the programme glossary from the existing controlled vocabulary. Every 504-covered organisation already has a programme-name list, an acronym handbook, and a regulatory citation index. The university registrar maintains the institutional name list; the hospital's GME office maintains the procedure and drug list; the school district's HR office maintains the federal-programme acronym list. Connect via Notion, Confluence, SharePoint, or Google Docs — or paste a flat list once.
Process the back-catalogue first. 504 enforcement looks backward. A complaint filed today reaches whatever videos are live today, regardless of when they were captioned. The remediation order in a resolution agreement typically requires re-captioning the existing catalogue, not just future content. GlossCap's batch flow processes a back-catalogue at the throughput cap of the host platform's upload API.
Reviewer pass before LMS upload. The amber-highlight UI shows every glossary-applied term with its source line, so a 504 coordinator (or department training lead) can scrub the high-meaning surface in minutes per video. Corrections feed the workspace glossary, so the same mangling doesn't re-occur.
Document the workflow for the resolution-agreement file. If OCR comes asking, the recipient that can show a captioning policy, vendor SLA, glossary source, and reviewer signoff for each asset moves through the investigation in weeks rather than quarters. GlossCap's per-asset audit log is designed to drop directly into a 504 documentation package.

See pricing

504, ADA, and Section 1557: when civil-rights regimes stack

A federally funded healthcare or higher-education organisation is rarely covered by 504 alone. The stack typically reads:

Section 504 — federal financial assistance trigger. OCR (Education) or OCR (HHS) is the enforcer.
Section 1557 of the Affordable Care Act — applies to any healthcare programme or activity that receives federal financial assistance, with broader scope than 504 in some respects (it cross-references Title VI, Title IX, the Age Act, and 504 itself). The 2024 final rule (89 FR 37522) explicitly reaches "language assistance services and effective communication," which OCR HHS reads to include captioning for video used in healthcare training and patient education.
ADA Title II — for state and local government entities, including state hospitals and public universities. WCAG 2.1 AA explicit since 2026-04-24 for large entities; see the ADA Title II captions reference.
ADA Title III — for private healthcare providers, private universities, and other places of public accommodation. WCAG 2.0/2.1 AA via case law.
Section 508 — where the entity also procures technology under a federal contract.

Resolving a complaint typically requires complying with the most stringent regime in the stack. For training-video captions, the most stringent regime is currently ADA Title II's WCAG 2.1 AA; meeting that bar resolves the 504 functional-access question by construction.