Four approaches changing cell therapy in 2026 — and how far each actually is

The field is usually described by what it is about to do. Induced pluripotent cells will make therapy available off the shelf. Gene editing will remove rejection. Organoids will replace animal testing. AI will fix manufacturing. Each of those is a fair description of a direction and a poor description of where things stand today, and you can measure the difference.

On 2026-09-05 we counted every study registered on ClinicalTrials.gov for each of these approaches, along with how many are recruiting and how many reached Phase 3. The queries are published below so the count can be repeated. What follows is those four approaches, each with the number attached and a source that is not us.

1. Induced pluripotent cells as an editable platform

The argument for iPSCs is not that they are more potent. It is that they are editable and inexhaustible. A single line can be expanded indefinitely, banked, and modified before anyone is treated — which turns a biological problem into a manufacturing one.

A 2025 review in Stem Cell Reports sets out both halves honestly. iPSCs offer “an essentially limitless source of cells, capable of differentiation into various immune cell types” and allow HLA-matched or engineered universal donor banks. The obstacles named are just as concrete: the endogenous T-cell receptor has to be disrupted by CRISPR-Cas9 or TALEN editing to avoid alloreactivity, HLA deletion still leaves cells “susceptible to clearance by innate immune effectors, particularly NK cells”, and efficient differentiation into functional T cells plus tumorigenicity risk remain unsolved.

The registry shows how early this still is. 172 registered studies, 55 recruiting, 2 at Phase 3. The underlying discovery won a Nobel Prize in 2012. Thirteen years later there are two late-phase studies, which says something about how hard the manufacturing half has turned out to be.

Reference: Fang Y, Chen Y, Li Y-R. Engineering the next generation of allogeneic CAR cells: iPSCs as a scalable and editable platform. Stem Cell Reports, 5 June 2025.

2. Off-the-shelf allogeneic products

Autologous therapy — cells taken from the patient, modified, returned — works but does not scale. Each dose is a bespoke manufacturing run with its own release testing, and the patient waits. The allogeneic answer is to make one batch from a donor and treat many people from it.

The registry number here is worth stating plainly. 274 registered allogeneic CAR studies. 111 recruiting. 0 at Phase 3. Not few — none. All 274 are still working on safety or early activity.

That does not mean the approach fails. It means the field is younger than the marketing suggests, and anyone selling an allogeneic product today is running ahead of the late-phase evidence.

3. Organoids replacing animal models

This one moved from research idea to regulatory position in a single announcement. On 10 April 2025 the FDA published a plan to phase out the animal-testing requirement for monoclonal antibodies and other drugs, promoting instead what it calls New Approach Methodologies: lab-grown human organoids, organ-on-a-chip systems, computational modelling and AI-based prediction. The agency's stated intent is to make animal testing “the exception rather than the norm” in preclinical safety within three to five years.

For stem-cell science this matters more than it first appears. Organoids are built from stem cells, so a regulatory push toward organoid-based safety testing is a regulatory push toward stem-cell-derived tissue as an industrial input rather than a treatment. The registry records 227 studies involving organoids or organ-on-chip systems, most of them using the tissue as a model rather than as a therapy.

Reference: U.S. Food and Drug Administration. FDA Announces Plan to Phase Out Animal Testing Requirement for Monoclonal Antibodies and Other Drugs, 10 April 2025.

4. Artificial intelligence in manufacturing

The fourth claim is the one most often made and least often examined. AI will not identify a new cell type; what it is actually being asked to do is hold a manufacturing process steady and predict whether a batch will work.

The second half of that runs into a wall that has nothing to do with computing. In its August 2026 draft guidance on potency for active immunotherapy products, the FDA states that assessing potency is difficult precisely because it depends on the host immune response. There is no validated assay that predicts clinical effect for most cell products — which means there is no reliable label for a model to learn from. A model can only be as good as the measurement it is trained against, and that measurement is the open problem.

The regulator's framing for AI itself is narrower than the marketing. The FDA's January 2025 draft guidance sets a risk-based credibility framework tied to a context of use: the sponsor states what question the model answers and what decision rests on it, and the evidence required scales with the consequence. A model scheduling a chiller and a model predicting batch release are the same algorithm under completely different obligations.

The comparison is easier to see next to a process that already works. Plasma fractionation has been running at scale since the 1940s and every batch gets an assay. Setting the two next to each other shows what AI is doing in each: in plasma it improves a process that is already measured, in cell therapy it is being asked to measure the thing first.

Further reading: AI in biologics manufacturing 2026: plasma and cell therapy compared

References: U.S. Food and Drug Administration, Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products (draft, January 2025); Regulatory Affairs Professionals Society, FDA drafts guidance on assessing potency of immunotherapy products, August 2026.

The pattern across all four

None of these is a discovery about biology. iPSCs fix supply. Allogeneic products fix scheduling and cost. Organoids fix the model problem in safety testing. AI is aimed at process control and batch release. All four are attempts to turn a craft into manufacturing, and all four run into the same missing piece: no reliable way to say whether a batch of cells will do anything.

That is worth keeping in mind when reading announcements over the next few years. Ask which problem the news is about — the biology, or making and measuring it. Nearly everything genuinely new in 2026 is the second kind, and there is nothing wrong with that. It is where the work is.

Repeat the count yourself

Every number above comes from one extraction on 2026-09-05. The exact API queries are below; running them will return today's counts, which will differ from ours as the registry changes.

Counts are our own extraction from the ClinicalTrials.gov API v2. A registered study is a declaration of intent, not a result: none of these numbers is evidence that any approach works, and nothing here is medical advice or a treatment recommendation. Cell-type detail behind these totals is on our cell type pages.

Sources & further reading

• Stem-cell trials registry (ClinicalTrials.gov) ↗
• Peer-reviewed research (PubMed) ↗
• ISSCR — patient resources ↗
• FDA — consumer guidance ↗
• EMA — ATMP framework ↗

Commercial price observations are heterogeneous and most cell-therapy uses remain investigational. Verify source scope, regulatory status and the proposed care with an independent qualified physician.