On 22 September 2026 the FDA deleted the word "animal" from five parts of the Code of Federal Regulations. Parts 312, 314, 315, 361 and 601 — the rules governing investigational drugs, marketing applications, biologics licences and safety reporting — now define a nonclinical test as any study run in vitro, in silico, in chemico, or in a living animal. Three non-animal examples are named in the definition itself: cell-based assays, organ chips and microphysiological systems, and computer modelling.

The agency's own accounting of what animal testing costs sits in a document from April 2025. A typical monoclonal antibody programme runs on 144 non-human primates. Each one now costs up to $50,000.

What the rule leaves open is how any one method earns the agency's trust. That question moved from policy to evidence in March 2026, and it is now being answered organ by organ.

🎯
The September rule changed the vocabulary of drug safety testing, not its evidentiary bar. A sponsor now argues that a method is fit for a stated purpose.

Emulate's Human Liver-Chip has filed a full qualification package: 27 known drugs, 87% sensitivity and 100% specificity for liver injury.

Xenotransplantation sits outside this framework entirely. There, the animal is the method.

Both numbers in that second line come from Emulate, in a September 2026 press release and in a webinar the company ran alongside the Physician's Committee for Responsible Medicine. The FDA has accepted none of it. The qualification is still pending.


Four sentences, and a shift in who has to prove what

Read the direct final rule and it does very little. FDA states plainly that it does not require a particular method and does not change the agency's evidentiary standards. The document is Docket FDA-2026-N-5347, RIN 0910-AJ27, and it substitutes terminology across five CFR parts.

5 CFR parts rewritten

Scope of the terminology rule

Parts 312, 314, 315, 361 and 601, under Docket FDA-2026-N-5347 and RIN 0910-AJ27 · Federal Register, 2026

That modesty is the whole strategy. A terminology rule relocates the cost of a dispute. Before September, a sponsor filing an IND with only human-cell data was arguing that its data counted at all. After September, the same sponsor is arguing that the data is fit for one stated use. The first is existential. The second is technical, and technical arguments get settled with validation studies and a comparator, not with a statute.

The three things a NAM has to demonstrate

FDA's draft guidance of 18 March 2026 — General Considerations for the Use of New Approach Methodologies in Drug Development — supplies that standard. It is deliberately not a list of approved methods, and it says so. The guidance does not address specific NAMs and does not address their use in discovery. It addresses submissions.

What a NAM must show FDA

Context of use. The narrow question the method is being asked to answer, stated in advance. A liver chip qualified for drug-induced liver injury in one chemical class transfers poorly to a cardiac safety question.

Biological relevance. Why human tissue carries information the previous model did not, in terms of the mechanism under investigation.

Technical characterization. Whether the system is stable, reproducible and characterised well enough for a reviewer to weigh its output against animal data.

Context of use is the hardest of the three, because it is the one that cannot be expanded later. CDER has run a live inventory of acceptable streamlined nonclinical approaches since October 2025. It currently lists a small set of contexts, including human induced pluripotent stem cell-derived cardiomyocytes for safety pharmacology and in vitro liver models for hepatotoxicity. That list is the practical measure of how far the framework has travelled.

PathwayWhat it costsWhat it settles
Non-animal NAM ◐ Validation studies ✔ One named context of use
Animal study ✗ Up to 144 primates per antibody programme ✔ Whole-organ toxicology, species-approximate
Weight of evidence ◐ Mostly existing human data ◐ Programme-level, reviewer discretion

Comparison drawn from FDA's April 2025 roadmap, the March 2026 draft guidance and the CDER streamlined nonclinical inventory · 2025–2026

Two FDA pathways do the qualifying. The Innovative Science and Technology Approaches for New Drugs program, ISTAND, and the Drug Development Tool qualification programme each take a method through to a status that can be cited in later filings. Both were named in the direct final rule itself.

Where the framework already has traction

The phase-down is running programme by programme, and it is furthest along where the animal study was weakest. Three areas have moved past argument.

Cardiac safety is the clearest. Human induced pluripotent stem cell-derived cardiomyocytes are now a new approach methodology used in regulatory submissions to the FDA, and the agency encourages sponsors to submit these studies where they follow best practice, so that reviewers can weigh how clinically relevant the findings are. That placement matters. Pro-arrhythmic risk is a question about human ion channels, and a human cardiomyocyte assay reads that mechanism directly.

Liver injury is the second. CDER's streamlined inventory lists in vitro liver models for assessing hepatotoxicity, built on work that measures liver injury through changes in biomarkers and functional endpoints. The citation in the agency's own table runs to 2023, which dates the evidence without dating the policy: the guidance stacked on top of it arrived in March 2026.

Manufacturing quality control is the third, and the least discussed. FDA's updated pyrogen and endotoxins guidance supports replacement of animal-based pyrogen tests such as the rabbit test with non-animal methods, including recombinant reagent-based endotoxin assays. These are high-throughput assays with no primate cost attached. A programme that still runs the rabbit test out of habit is buying very little.

The pattern across all three is worth stating directly. The framework moves fastest where a human alternative already produces a quantitative readout mapped to a question the FDA asks anyway. It stalls where the animal stood in for a whole organism. A single-organ assay does not cover that.

The liver chip that got furthest

Emulate submitted its Full Qualification Package to ISTAND on 9 September 2026, covering both the Zoë culture module and the newer AVA emulation system. The supporting study evaluated 27 small molecules with known clinical outcomes and reported 87% sensitivity with 100% specificity for drug-induced liver injury liability, including severe injury that animal testing had not reliably flagged.

87% sensitivity, 27 drugs

Human Liver-Chip DILI detection

Reported with 100% specificity on the same panel; qualification package filed with FDA ISTAND · Emulate, 2026

Those are sponsor numbers. Twenty-seven compounds is a small panel, and specificity is measured against a labelled ground truth that human clinical data itself approximates. Qualification is a determination the FDA has not yet made.

⚠️
The problem the chip is aimed at
Drug-induced liver injury stays one of the hardest safety questions in development. Candidates that look clean in conventional models still injure livers in patients, and the failure surfaces as trial attrition, regulatory setbacks and withdrawals after a drug is already on the market. If a human liver system can catch what the animal study missed, the economic case writes itself — provided the regulator accepts the data.

Whether it does is a separate question from whether it works. Emulate has been in ISTAND before: the Liver-Chip S1 was accepted into the pilot program, and the company has an existing co-operative research agreement with FDA to run cross-species toxicology comparisons on its human, dog and rat liver chips.

The arithmetic that started it

FDA's April 2025 roadmap began the phase-down with monoclonal antibodies, where the long-term primate study is expensive, slow and, in the agency's framing, poorly predictive of human outcomes. The 2025 draft guidance recommends eliminating or reducing long-term primate toxicology for certain antibodies, using computational models and human-relevant systems under a weight-of-evidence assessment. A pyrogen and endotoxins guidance update in 2026 goes further, supporting replacement of the rabbit test with recombinant reagent-based assays in manufacturing quality control.

144 primates per programme

Animal cost, typical mAb programme

At up to $50,000 per non-human primate, as cited in FDA's roadmap · FDA, 2025

None of that requires a chip. A weight-of-evidence case can be assembled from human data that already exists. The organ-on-chip platform earns its place only where the animal study was carrying the entire safety argument and a human system can carry it better.

As we wrote in July, on organ chips moving from discovery into drug development, the binding constraint was never the hardware. It is whether a method can be shown fit for one named purpose, in one indication, on evidence a regulator accepts. Emulate's Liver-Chip S1 entering the ISTAND pilot programme was the first candidate to sit that test down.

The field the framework does not reach

Every method above assumes a human body is available for measurement. Xenotransplantation removes it. A gene-edited pig kidney is raised in an animal, assessed in an animal, and its safety margin established in animal studies, because that is the only model in existence.

The programme's progress is real. eGenesis reported in September 2026 that five patients had received its EGEN-2784 porcine kidney under FDA expanded access since March 2024, three with sustained kidney function for at least eight months without dialysis, and two who had since transitioned to a human donor kidney. The RESTORE trial — Phase 1/2/3, 33 patients aged 50 to 70 already waitlisted — is FDA-cleared, with initiation expected in the first quarter of 2027.

None of that evidence can be produced on a chip. Joshua Mezrich, a transplant surgeon at the University of Wisconsin, put the tension plainly in STAT in June 2026, arguing that retiring animal models would undercut the one field whose entire translational bridge is the animal. The FDA's framework has no category for this. It never needed one, until now.

📊
Signals to track through 2027

How many contexts of use CDER adds to its streamlined nonclinical inventory — it went live in October 2025 with a short list.

How many organ chips clear ISTAND or Drug Development Tool qualification. Each qualification lowers the cost of arguing the next one.

Whether the 2027 mAb guidance lands as a rule or stays a draft. Drafts can be declined.

Until an organ system earns qualification on evidence the FDA itself accepts, the September rule changes what a sponsor must argue. It does not change what a sponsor can prove.

New Approach Methodologies — FDA programme page
The agency's running index of NAM work: the September 2026 terminology rule, the March 2026 validation draft guidance, the 2025 antibody guidance, and the ISTAND and qualification programmes.
Primary source. Every regulatory date in this piece traces back to this page or the documents it lists.
Nonclinical Testing Terminology — proposed rule, 91 FR, 22 September 2026
The rule text itself, including the definition of "nonclinical test" and its three named non-animal examples. Companion to the direct final rule published in the same issue.
Where the statutory wording lives. FDORA 2022 section 3209(a) is what forced the definitional change in the first place.
Emulate files its Human Liver-Chip qualification package with FDA
The 9 September 2026 submission covering the Zoë and AVA platforms, and the source of the 27-drug DILI study reported at 87% sensitivity and 100% specificity.
Company source, and treated as such here. The performance figures are the sponsor's, not the regulator's.
What ending animal testing does to xenotransplantation
Joshua Mezrich argues that the retreat from animal models collides directly with the field still relying on large-animal studies to prove a pig organ is safe in a human.
The clearest statement of the collision this rule creates, from a clinician running the programmes rather than the platforms.