TL;DR: After decades as a lab curiosity, alternatives to animal testing have become a regulatory reality. In September 2026 the US FDA rewrote its drug-development vocabulary to say “nonclinical tests” instead of “animal tests”, following a 2022 law and a 2025 pledge to make animal studies the exception rather than the norm. The technology — organs-on-chips, organoids, computer models — now works well enough to argue about; what remains is the harder work of validation, standardisation and changing scientific habits.
Why alternatives to animal testing moved from theory to policy
For most of the last century, the question “does this molecule harm humans?” was answered by giving it to an animal and watching. That default was written into regulation, taught in laboratories and baked into careers. It is now coming apart, and the pace of the last four years is the story.
The first crack was legislative. Late in 2022, the US Congress passed the FDA Modernization Act 2.0, which explicitly authorised new approach methodologies — NAMs — in the preclinical studies a drug needs before it can be tried in people. Nothing was banned. But for the first time, the law said a company could bring human-cell systems, organ chips or computational models instead of a rodent study and expect to be taken seriously.
Then the regulator started acting like it meant it. In April 2025 the FDA published a roadmap pledging “to make animal studies the exception rather than the norm” for drug safety testing. In July 2025 the National Institutes of Health — the world’s largest public funder of biomedical research — told grant applicants that work on animal models would also need to incorporate non-animal research. The European Commission published a roadmap toward phasing out animal testing in chemical safety assessment, the UK announced its own plan built around AI and 3D bioprinting, and the OECD updated its influential test guidelines to allow wider use of NAMs.
The decisive step came on 21 September 2026, when the FDA issued a direct final rule on nonclinical testing terminology. It replaces “animal tests” and “animal studies” in the agency’s drug regulations with “nonclinical tests” and “nonclinical studies”, and replaces “preclinical” and “in vitro” with the same broader term. The definitions track the Food and Drug Omnibus Reform Act of 2022. In plain terms: the paperwork no longer implies that a cage is the only valid source of safety evidence.
Acting FDA Commissioner Kyle Diamantas framed it as flexibility rather than a ban. The rule, he said, “does not eliminate or prohibit animal studies” and does not change evidentiary standards or impose new costs. The agency also opened a database of 25 real NAM use cases drawn from its own review materials, so sponsors can see what “accepted” actually looks like. The rule carries a comment period; if the FDA receives significant adverse comments it will withdraw the direct final rule and proceed through the slower proposed rule instead. That procedural hinge matters — it means the shift is real but not yet nailed down.
The signal has already reached industry. Ilka Maschmeyer, a translational toxicology researcher at the German organ-chip company TissUse, describes a recent case in which a pharmaceutical company was refused permission by the FDA to start a clinical trial: it had submitted animal data, and the agency wanted organ-chip data or something comparable. “It’s rare still,” she says, “but I think it’s going to be more and more frequent.”
What “alternatives to animal testing” actually means
NAMs is an umbrella acronym — new approach methodologies, novel alternative methods, nonanimal methods, depending on who is talking. Under it sit four families of technology that are often confused with each other.
Organoids
Organoids are self-organising three-dimensional clusters of stem cells that partly reproduce the structure and function of a real organ: miniature guts, brains, livers, tumours. They are cheap enough for ordinary academic labs and good at capturing human cell biology that a mouse cannot. Their weakness is control — two organoid batches are rarely identical, which is fatal for a regulatory assay unless the process is tightly specified.
Organs-on-a-chip
An organ-on-a-chip is a slab of clear polymer the size of a USB stick, threaded with microscopic channels lined with living human cells, and perfused so the tissue experiences something like blood flow. The founding demonstration came from Donald Ingber’s team at Harvard’s Wyss Institute: a lung model whose cell-lined channels stretched and relaxed as air was pumped through neighbouring chambers. It breathed.
That paper had an instructive history. Science rejected the first submission and suggested the team also run the experiment in mice. Ingber did, resubmitted a year later, and the study was published in 2010 — it has since been cited nearly 5,400 times. The episode is a neat summary of how the field was treated: promising, but only credible once an animal agreed.
Since then the catalogue has widened to livers, hearts, kidneys, brains, placentas and vascularised tumours. As many as ten organ models have been linked into multi-organ systems that approximate human physiology well enough to trace how a drug is absorbed, metabolised in the liver and then hits a second tissue — the interaction that single-cell cultures and mice both handle badly.
Computational models and digital twins
The third family skips biology for mathematics: simulations of human heart cells, pharmacokinetic models, and “digital twins” in which chip data feeds a whole-body prediction of what a drug will do. This is where AI earns its place, not as a replacement for experiments but as the loop that designs the next one.
The combination is the point
Real programmes stack these. Chips generate human-relevant data; organoids supply donor diversity; models extrapolate to a dose in a person. Ingber’s lab has shown multi-organ chips predicting drug behaviour in humans; Christian Maass at the German firm ESQlabs builds digital twins fed by chip data. Individually each method is a partial answer. Together they are the first credible substitute for the whole-animal assay.
The evidence: where the alternatives already beat mice and monkeys
The strongest argument for NAMs is not ethical, it is numerical. An estimated 92 percent of drugs that enter US clinical trials fail to reach the market, often because they were ineffective or unsafe in ways animal experiments did not predict.
Two studies are usually cited. A liver-on-a-chip system from Emulate correctly flagged about seven out of every eight drugs that had passed animal trials but proved toxic to human livers. Separately, researchers from Oxford and Janssen Pharmaceutica showed that computational simulations of human heart cells predicted a dangerous arrhythmia with 89 percent accuracy, against 75 percent for the animal studies.
That is the shape of the case: not that animals are useless, but that for specific questions about human tissue, a human system can be measurably better. The same logic cuts both ways. Animal models can also wrongly condemn a drug that would have worked — some researchers argue aspirin or paracetamol, discovered today, might never have survived the testing gauntlet.
| Method | What it captures | Typical cost and throughput | Regulatory standing (2026) |
|---|---|---|---|
| Animal study | Whole-organism physiology, but of another species | High cost, low throughput, months | Still accepted; no longer the only route in US drug rules |
| 2D cell culture | One cell type, static, no mechanical forces | Very low cost, very high throughput | Accepted as supportive data |
| Organoid | 3D human tissue architecture and genetics | Moderate cost, medium throughput, batch variability | Accepted case by case, OECD guidelines widened |
| Organ-on-a-chip | Perfused human tissue with mechanical forces and multi-organ links | High cost per study, low throughput, needs skilled operators | Explicitly named in FDA’s September 2026 rule |
| In silico / digital twin | Predicted human response at scale | Low marginal cost once built, very high throughput | Accepted where validated for a defined question |
Validation is the bottleneck, not the biology
If the science works, why is a mouse still cheaper than a chip? Because “it works in the lab that invented it” is not a regulatory standard.
“You read about all the organ chips that come out of academic labs, which is great — but that’s not going to change their uptake by the FDA, because you have to get the same results anywhere in the world,” Ingber says. “It has to be a commercial product. It has to be mass-produced and meet very fine performance criteria.”
The failure modes are unglamorous. Minute differences in the hydrogels used as tissue scaffolds produce visibly different growth patterns. Different research groups measure blood-vessel function and geometry with incompatible metrics, so a tumour-on-a-chip result from one lab cannot be compared with another’s. Guidelines, workflows, checkpoints and reporting criteria all have to be agreed before the data means anything.
And a validation study is not cheap. Emulate’s liver-chip qualification needed 870 chips and the labour equivalent of 16 full-time staff for 16 weeks — beyond the reach of most academic labs. Extending the claim from small-molecule to large-molecule drugs would require another study of that size, and comparable work is ostensibly needed for every commercial NAM in every context of use. “It takes too long and costs too much, and small companies can’t afford to do it,” says Kathy Archibald of the UK’s Safer Medicines Trust.
This is the honest state of play in late 2026: the technology has outrun the evidence base. Maass of ESQlabs puts it bluntly — the field has not yet produced “the evidence and the proof that we are doing better or as good as the animal models.” He compares the missing moment to the iPhone launch, when people saw at once that a phone without buttons could work. “That moment is still lacking.”
The fixes are institutional: governments funding head-to-head comparisons, pharma and regulators opening proprietary toxicity data, and a push — backed by the FDA — for an open repository of drug toxicity data so that new methods can be checked against known human outcomes.
The culture problem: rules change faster than habits
Regulation can be rewritten in a year. Habits take a generation.
The monocyte activation test is the cautionary example. Developed and validated in the mid-1990s, it uses human blood cells to detect pyrogens — fever-causing contamination — and replaces the rabbit pyrogen test, in which a compound is injected into a rabbit’s ear and its rectal temperature monitored. The European Pharmacopoeia only accepted it as a replacement in 2010, roughly fifteen years later. Rabbits are still widely used for that test around the world.
“The formal requirement may disappear, but the informal expectation persists,” says Kathrin Herrmann, a veterinary scientist at Johns Hopkins’ Center for Alternatives to Animal Testing. Regulators, grant reviewers, peer reviewers and journal editors are the human infrastructure of science, and many of them are unfamiliar with NAMs and reflexively want animal data. Herrmann’s survey of early-career researchers in the field keeps turning up the same complaints: NAM-only proposals are read as risky, there is pressure to bolt on an animal experiment for credibility, infrastructure access is limited, and careers look uncertain for anyone who departs from established models.
There are practical frictions too. A toxicologist who has spent twenty years running rat studies is being asked to bet a programme on unfamiliar polymer and stem cells. An academic whose lab is built on animal models may see a switch as the end of jobs for people whose skills no longer apply. “I could see why it’s a hard thing for people to take it up,” Ingber concedes.
Training is where the field is placing its bets. Thomas Hartung’s Coursera modules on alternatives to animal testing have been taken by roughly 12,000 students. Joseph Wu, director of Stanford’s Cardiovascular Institute and co-founder of Greenstone Biosciences, has pushed his company to share its stem-cell line library freely with academic researchers. “These trainees will set up their own labs. They will go into industry. They will replace the old guard,” he says.
What this means for Malaysian labs, CROs and exporters
Malaysia does not write the rules that govern drug safety, but it works inside them — and the change of direction abroad lands here in three concrete places.
Contract research and university labs. A Malaysian CRO or university group that can run a qualified organ-chip or organoid study becomes subcontractable by sponsors who now need human-relevant data for a US filing. Buying a validated platform is expensive; offering a service on shared infrastructure is the cheaper entry point, and it is the same playbook the local cell-culture and bioanalytical labs already run.
Export-facing chemical and cosmetic testing. Malaysian oleochemical, surfactant and cosmetic ingredient producers test to OECD guidelines because that is what buyers in the EU and Japan require. The OECD’s expanded acceptance of NAMs, plus the NPRA’s own cosmetic safety assessment guideline — which notes that finished-product testing in almost all cases does not need animal toxicology, and that alternative methods should be ones generally accepted under OECD guidelines — reduces the cost of the irritancy and sensitisation work that sits behind an ASEAN Cosmetic Directive product notification.
Research governance. Locally, animal work is governed by the Animal Welfare Act 2015 (Act 772), in force since July 2017, and the Malaysian Code of Practice for the Care and Use of Animals for Scientific Purposes. Any use of live vertebrates needs Institutional Animal Care and Use Committee approval, and MYCODE already obliges investigators to justify animal use against the 3Rs: replacement, reduction, refinement. The practical consequence is that a Malaysian IACUC review will increasingly ask why a validated non-animal method was not used — the same question US and EU reviewers now ask, in a local legal frame.
There is a commercial read too. The organ-on-chip market was estimated at about US$186 million in 2025 and is forecast to reach roughly US$1.2 billion by 2033, with services the largest segment today. Malaysia’s realistic role is not manufacturing chips but supplying the biology-adjacent services around them: cell sourcing, assay development, data work.
One caveat worth stating plainly. Around 30 percent of animals used in experiments go to early-stage drug development and regulatory testing. The rest go to basic biological research, where replacement is much less straightforward — you cannot put an entire developing organism on a chip yet. The regulatory shift described here covers that 30 percent, and it does not automatically generalise.
FAQ
Do the FDA’s 2026 changes ban animal testing?
No. The September 2026 direct final rule only changes terminology in the agency’s drug regulations, replacing “animal tests” with “nonclinical tests” and clarifying that validated non-animal methods can be used where appropriate. The FDA states explicitly that it does not eliminate or prohibit animal studies, does not change evidentiary standards and does not impose new costs.
Are organ-on-a-chip results accepted for a new drug application today?
Accepted when the method is adequately validated and appropriate for the specific product and regulatory question — not automatically. The FDA has published a database of 25 use cases from its own review materials showing where NAMs have been relied on. Emulate’s liver chip has been qualified for drug-induced liver injury assessment; other claims need their own validation studies.
How much does a validation study cost?
There is no public price list, but the labour gives a sense of scale. Emulate’s liver-chip qualification used 870 chips and the equivalent of 16 full-time employees for 16 weeks, and a further study of similar size would be needed to extend the claim to large-molecule drugs.
Can Malaysian companies file with NAM data?
The rule that matters is the one at the destination regulator. A Malaysian sponsor filing with the US FDA or the EMA can present NAM data under the same conditions as any other sponsor; what is scarce locally is qualified infrastructure and trained operators, which is why service arrangements with established chip providers are common.
Will animal testing disappear in the next decade?
Probably not entirely. The regulatory direction is clearly away from animal studies as a default, and the 2022 US law, the 2025 FDA roadmap and the September 2026 rule all point the same way. But validation backlogs, publishing norms and training pipelines move slowly, and basic research — the majority of animal use — has no near-term replacement.
The bottom line
The interesting thing about the current moment is not that a chip can imitate a lung. Ingber showed that in 2010. It is that the institutions around the technology have started to catch up: a 2022 law that permitted NAMs, a 2025 pledge to make animal studies exceptional, a 2026 rule that rewrote the vocabulary, and parallel moves from the EU, the UK and the OECD.
What is left is the unglamorous half — standardised protocols, head-to-head evidence, shared data, trained people, and reviewers who no longer flinch at a submission without a rodent arm. Thomas Hartung calls the toxicology transition a “lighthouse function” for the rest of biology. “Suddenly,” he says, “all the dams have opened.” For Malaysian labs, the practical question is narrower and answerable this quarter: which validated non-animal assay can we run, and for whom?