CRISPR Gene Editing in 2026: What It Can and Cannot Do

What CRISPR Gene Editing Is, and What It Can Treat in 2026

TL;DR: CRISPR-Cas9 does not edit DNA. It cuts DNA, and the cell repairs the cut — that distinction explains almost everything else, including why the newer editors were invented to avoid the cut altogether. As of August 2026, one CRISPR therapy holds marketing authorisation from a major regulator, for two blood disorders. The first in vivo phase 3 trial succeeded in April 2026. An infant received a therapy designed for his private mutation in six months. And a patient died in a CRISPR trial in November 2025. The bottleneck is no longer whether we can edit — it is getting the machinery into the right cells of the right tissue. That single constraint explains why the approved treatments target blood and liver, and not brain or muscle.

This article is general educational information about a field of research. It is not medical advice, not a recommendation for or against any treatment, and not promotional material for any medicine. Treatment decisions belong with a qualified clinician.

What CRISPR Actually Is

CRISPR stands for Clustered Regularly Interspaced Short Palindromic Repeats — a description of a pattern in bacterial genomes, which tells you nothing useful about what it does.

What it does is bacterial immune memory. In 2007, researchers studying Streptococcus thermophilus — the bacterium used industrially to make yoghurt and cheese — showed that after a virus attack, surviving bacteria filed away fragments of the viral DNA in their own genome. Those stored fragments, called spacers, let them recognise and destroy the same virus next time. It was heritable, sequence-specific immunity (Barrangou et al., Science, 2007).

The discovery came out of an industrial problem — bacteriophages ruin fermentation batches — not a gene therapy lab. What made it revolutionary was the realisation that the targeting system was programmable.

The European origin story that usually gets skipped

The repeats themselves were first described in 1993 by Francisco Mojica at the University of Alicante, working on salt-loving archaea. Mojica spent the following decade establishing what they were: in 2002 he proposed the acronym CRISPR itself, and in 2005 he showed that the stored spacers come from foreign DNA — phages and plasmids — and proposed that the whole system was a prokaryotic adaptive immune system (Mojica et al., J Mol Evol, 2005).

In 2009 he described the PAM, the short motif that determines which sequences the system can target at all (Mojica et al., Microbiology, 2009). That detail matters for the section below, and it is routinely misdated to 2005 in popular accounts.

How CRISPR-Cas9 Works, Step by Step

In 2012, Jennifer Doudna and Emmanuelle Charpentier's groups showed that the two RNA molecules bacteria use to guide the Cas9 enzyme could be fused into a single guide RNA — a programmable ~20-base sequence that directs Cas9 to any matching stretch of DNA (Jinek et al., Science, 2012). They shared the 2020 Nobel Prize in Chemistry for it (NobelPrize.org).

The mechanism has three parts:

1. The guide RNA finds the target. It base-pairs with the matching DNA sequence — the same principle of complementary base pairing that underlies how DNA testing reads your genome.
2. The PAM licenses the cut. Cas9 cannot cut just anywhere: it needs that short motif sitting next to the target. For the standard Cas9, it is NGG — common enough across genomes not to be a serious practical limit. The PAM is also why bacteria don't destroy themselves: their own stored spacers lack an adjacent PAM.
3. Two nuclease domains cut. HNH and RuvC each sever one strand, producing a double-strand break.

And then Cas9's job is finished.

The part most explainers get wrong

CRISPR-Cas9 does not edit anything. It cuts. The cell does the editing — and the cell has two very different ways of repairing a double-strand break.

This is the asymmetry that shapes the entire field: switching a gene off is easy; correcting one is hard. Every technology in the "beyond Cas9" section below exists to get around it.

Reading DNA Versus Writing DNA

There is a confusion worth clearing up early, because it sits underneath a lot of misplaced hope.

Sequencing or genotyping your DNA is a reading operation. It tells you what your genome says. Gene editing is a writing operation. It changes what your genome says. They are separate technologies with separate infrastructure, and one does not lead to the other.

There is no pathway today from "I have my genetic data" to "therefore I can have it edited." If you ever needed a CRISPR therapy, the starting point would not be a consumer DNA file — it would be confirmatory clinical sequencing ordered by a physician. Consumer genotyping arrays check a pre-selected set of common positions; they are not diagnostic, and they do not identify the mutations these therapies correct.

Worth adding: gene editing is also not the same thing as gene therapy in the classical sense. Traditional gene therapy adds a working copy of a gene, leaving the broken original in place. CRISPR modifies the existing sequence. The terms are not interchangeable.

Ex Vivo Versus In Vivo: The Distinction That Decides What Is Treatable

This is the most useful concept in the field.

Ex vivo editing means the cells are removed from the patient, edited in a dish where conditions are controlled, checked, and then returned. You can verify the edit worked before putting anything back. The catch is that it only works for tissues you can actually take out and put back: blood stem cells and immune cells. You cannot do this to a liver.

In vivo editing means the editing machinery is injected into the patient and has to find the right cells on its own. It opens up solid organs — and it means you cannot inspect the result before it happens.

Nearly everything approved or far along in trials today is either blood (because it can be edited ex vivo) or liver (because the delivery vehicles naturally accumulate there). That is not a coincidence about which diseases matter most. It is a map of what delivery can currently reach.

Why delivery is the real bottleneck

A 2026 review from Doudna's group puts it directly: "The difficulty of tissue-specific delivery currently limits applications of editing technology" (Ngo et al., Nature Biotechnology, 2026).

There are three routes, and each has a hard limit:

• AAV viral vectors carry a genetic payload with good tissue targeting, but their packaging capacity is roughly 4.7 kilobases — the size of the virus's own genome — with a practical ceiling around 5 kb. Cas9 plus its guide barely fits, and the larger editors described below do not fit at all. Much of the population also has pre-existing immunity to the viral capsid, and most serotypes drift to the liver anyway.
• Lipid nanoparticles — the same class of technology behind mRNA vaccines — are non-viral, can be re-dosed, and produce only transient expression, which reduces off-target risk. But they are cleared by the liver, and getting them anywhere else is the major unsolved problem.
• Ex vivo electroporation gives the highest efficiency and full quality control, but only for cells you can remove and reinfuse.

The tissues that remain hard to reach: the central nervous system, skeletal muscle, and several immune cell populations. Until that changes, the list of treatable conditions stays shaped more by delivery than by biology.

What CRISPR Can Treat Right Now

As of August 2026, and to the best of our knowledge, one CRISPR-based therapy holds marketing authorisation from a major regulator.

Casgevy (exagamglogene autotemcel) treats sickle cell disease and transfusion-dependent beta-thalassemia. It is an ex vivo therapy: the patient's own blood stem cells are removed, edited to switch fetal haemoglobin back on, and reinfused.

The reported long-term follow-up data are as follows. In the manufacturer's December 2025 data cut, 45 of 45 evaluable sickle cell patients aged 12 and over met the primary endpoint of at least 12 consecutive months free of vaso-occlusive crises, with a mean duration of 35.3 months (Vertex, ASH 2025).

The European authorisation is conditional, and carries an obligation to track treated patients in a registry for 15 years. That is the regulator saying, in effect: this works well enough to release now, and we do not yet know the long-run story.

Approval is not access

Casgevy's US list price is $2.2 million. In the UK it is £1.65 million, with a confidential discount and a managed access agreement; NHS England reached reimbursement deals for beta-thalassemia in August 2024 and sickle cell in January 2025. Italy's medicines agency agreed terms in September 2025, and Germany's statutory insurers in May 2026.

But price is not the binding constraint. According to reporting by STAT News, by February 2026 — more than two years after approval — roughly 60 patients had completed treatment worldwide, while around 147 had begun cell collection during 2025 (STAT, February 2026). Reported bottlenecks include the limited number of authorised treatment centres and difficulty collecting sufficient stem cells at several sites.

The Clinical Pipeline

Three developments in 2025–2026 matter more than the rest.

The first successful in vivo phase 3 trial. Intellia's lonvoguran ziclumeran, for hereditary angioedema, reported positive phase 3 results on 27 April 2026: an 87% reduction in attacks versus placebo across 80 patients (Intellia). It holds PRIME designation from the EMA. It remains investigational and is not authorised by any regulator. This is the first time editing performed inside the body has cleared a phase 3 endpoint.

Base editing for cholesterol. Verve's VERVE-102 switches off PCSK9 in the liver with a single infusion. Phase 1b interim results in May 2026 across 35 patients showed LDL cholesterol reductions of up to 62% at the highest dose, with follow-up extending to 18 months in the earliest cohorts and a median follow-up of roughly nine months (Eli Lilly). It remains investigational, is not authorised by any regulator, and has not been evaluated in a phase 3 outcomes trial. Note also what it targets: a single gene regulating cholesterol — a risk factor — rather than heart disease itself. It sits alongside the more established practice of using genetics to predict how patients respond to drugs.

CRISPR-edited CAR-T cells for B-cell cancers have produced complete responses across dose levels, and the approach is now being extended to autoimmune disease.

Beyond Cas9: Base Editing and Prime Editing

Both technologies exist to avoid the double-strand break, because the break is what causes most of the collateral damage.

Base editing (David Liu's lab, 2016–2017) fuses a disabled Cas9 to an enzyme that chemically converts one DNA letter into another, directly, without cutting both strands (Gaudelli et al., Nature, 2017). It handles the four "transition" changes — which, per Rees and Liu, covers roughly 61% of pathogenic point mutations catalogued in ClinVar as of 2018. It cannot do the other substitutions, and it cannot insert or delete.

Prime editing (Anzalone et al., 2019) fuses a nicking Cas9 to a reverse transcriptase, guided by an RNA that both specifies the target and encodes the desired edit — search and replace, with no double-strand break and no donor DNA template. The authors write that it "in principle could correct up to 89% of known genetic variants associated with human diseases" (Nature, 2019).

The useful mental model: Cas9 breaks and hopes. Base editing rewrites one letter. Prime editing searches and replaces.

Read those percentages carefully. Both are theoretical scope — which mutation types the chemistry can address, on a specific database at a specific date. The authors' hedges ("in principle", "up to") are doing real work. Neither figure says anything about whether the editor can be delivered to the right tissue, at sufficient efficiency, safely.

They are no longer purely theoretical, though. In 2025, an infant treated at the Children's Hospital of Philadelphia, born with a severe CPS1 deficiency, received a base editor designed specifically for his own private mutation, delivered by lipid nanoparticle. It went from design to manufacture to dosing in about six months. A year later he had avoided a liver transplant and was tolerating a more normal diet (CHOP). This is a single patient, still under long-term monitoring. It demonstrates feasibility, not established efficacy or safety, and the approach has not been evaluated in a controlled trial.

The regulatory consequence may matter more than the case itself. In February 2026 the FDA published draft guidance creating a "plausible mechanism" pathway, under which a single trial can evaluate a personalised editing platform across patients with different mutations, rather than treating each bespoke therapy as a separate drug. For ultra-rare disease, that changes the economics of the entire category.

Beyond Medicine: Diagnostics, Agriculture and the Lab

Most CRISPR coverage is about therapy. Most CRISPR use is not.

Not every Cas is Cas9. Cas12a cuts double-stranded DNA but needs only a single guide RNA and leaves staggered ends. Cas13 targets RNA rather than DNA — which makes its interventions transient and non-heritable, a meaningfully different risk profile.

A quirk of both became a diagnostic platform. Once Cas12a and Cas13 find their target, they start cutting nearby single-stranded nucleic acids indiscriminately. That "defect" is the basis of SHERLOCK (Science, 2017) and DETECTR (Science, 2018): add a reporter probe, and chopping it up produces a signal. Target present, signal on — at attomolar sensitivity, without a laboratory.

In agriculture, Europe has just changed its rules. Regulation (EU) 2026/1388 of 17 June 2026 entered into force on 16 July 2026, with full application after a two-year transition (EUR-Lex). It splits gene-edited plants into two tiers: NGT-1, where the changes could have arisen through conventional breeding, handled essentially as ordinary crops; and NGT-2, which stays under the existing GMO regime with full authorisation and labelling. The underlying shift is that plants are now judged by the genetic characteristics of the final product rather than by the method used to produce it. Note the scope carefully — this regulation covers plants. It has nothing to say about human therapy or genetic data.

And the largest use of all is unglamorous: CRISPR as a laboratory tool, systematically switching off genes one at a time to find out what each one does. That is quietly the technology's biggest contribution to biology.

Where AI enters

Designing a guide RNA, predicting where else in the genome it might bind, and engineering better nucleases are all pattern problems on sequence data — which is why machine learning has moved into each of them. Protein structure and sequence models are now used to propose novel editing enzymes rather than only catalogue natural ones. If you want that thread in depth, we covered protein language models and the wider field of AI in genomics separately.

It fits this article's argument: the design step is no longer the hard part. Delivery is.

What CRISPR Cannot Do

The limits are as informative as the capabilities, and they are structural rather than temporary.

It cannot treat polygenic conditions. CRISPR corrects single, well-defined mutations. Type 2 diabetes, heart disease, most cancers, depression, obesity — these arise from thousands of variants each contributing a fraction of the risk, interacting with environment and behaviour. There is no single edit to make.

It cannot reach most tissues. Brain, muscle, and much of the immune system remain out of reach for systemic in vivo editing.

It cannot deliver the traits people imagine. Intelligence, height and athletic ability are massively polygenic — height alone involves on the order of tens of thousands of variants. "Designer babies" in the popular sense are less a near-term ethical dilemma than a biological misunderstanding.

It cannot edit your whole body. Approved therapy edits one cell type. Your genome is not rewritten; a specific population of cells is.

It is not 100% precise. See the safety section below — on-target structural damage is documented, and some off-target effects are invisible to the algorithms meant to predict them.

It is not available to buy. Access means either an approved therapy for a specific indication, or a clinical trial with strict inclusion criteria. There is no clinic where a healthy person can have a variant "fixed."

And it will not cure cancer or Alzheimer's soon. Those are multifactorial. The real pipeline is monogenic disease plus cell-based immunotherapy.

The Safety Question

At least two patients have died in trials of CRISPR-based therapies, and neither death was attributed to the editing itself. Both are worth understanding. (Counts vary across coverage depending on which gene therapy platforms are included.)

In Intellia's ATTR amyloidosis programme, a man in his eighties dosed on 30 September 2025 developed grade 4 liver enzyme elevations, and died on 5 November 2025. The FDA placed both phase 3 trials on clinical hold on 29 October. Intellia later attributed the death to sepsis following an abdominal perforation found in surgery, and the FDA lifted the hold on one trial in January 2026 with enhanced liver monitoring (Intellia).

In a base editing trial for sickle cell disease, a patient died of respiratory failure four months after treatment. Investigators, an independent safety board and the FDA attributed it to the busulfan conditioning regimen, not the edited cells.

That second case points at something important and widely missed: in ex vivo therapies, including the approved one, the dominant documented risk is not the editing — it is the chemotherapy required to clear the bone marrow first. Infertility, lung toxicity and secondary malignancy risk come with that conditioning, not with CRISPR.

On the molecular side, the documented problems are real and remain active research areas:

• Large deletions and complex rearrangements at the target site itself, not just off-target (Kosicki et al., Nature Biotechnology, 2018).
• A p53 damage response to the double-strand break, which means editing preferentially selects for cells with a compromised p53 pathway — a theoretical cancer risk (Haapaniemi et al., Nature Medicine, 2018).
• Structural changes invisible to prediction algorithms: chromosomal deletions have been found at off-target sites with no sequence similarity to the guide RNA at all (Nature Communications, 2023). A 2025 review surveys the wider problem of structural variants and genome integrity (Nature Communications, 2025).

CRISPR in Europe

Two things make the European position distinct.

Editing that would be inherited is prohibited. The Oviedo Convention — the only binding international treaty on human rights in biomedicine — states in Article 13 that genome modification may be undertaken only for preventive, diagnostic or therapeutic purposes, "and only if its aim is not to introduce any modification in the genome of any descendants" (Council of Europe). A 2021 re-examination concluded the article needs no amendment. The EU Clinical Trials Regulation reinforces this independently, prohibiting gene therapy trials that result in germline modification.

This distinction — somatic editing, which affects only the treated person, versus germline editing, which would be passed to their children — is the single most important line in the field, and the one the public most often misses. Every approved and trialled therapy discussed in this article is somatic.

Trials face an extra layer. Advanced therapies containing genetically modified organisms must satisfy EU GMO legislation, transposed differently in each member state, before a clinical trial can begin — a documented source of delay for European trials relative to other regions. Europe's tendency to regulate genetic material more tightly than other jurisdictions is a pattern that also shows up in how genetic data itself is protected under GDPR.

For context on how far outside that consensus the alternative sits: in 2018 He Jiankui announced the birth of gene-edited twins in China. A Shenzhen court found he had falsified ethics documents and sentenced him to three years in prison; he was released in 2022. The case triggered tighter oversight of germline editing internationally. Meanwhile, US startups announced in 2025 that they intend to pursue heritable embryo editing — while scientific societies have called for a global moratorium, and a US federal budget provision bars the FDA from even considering such applications.

Where Your Own DNA Data Fits In

Mostly, it doesn't — and it is worth being clear about that rather than blurring it.

There is a real conceptual relationship, though, and it runs the opposite way to the one people expect. CRISPR today treats monogenic conditions: one gene, one mutation, one severe outcome — sickle cell disease, beta-thalassemia, ATTR amyloidosis. Almost everything a consumer DNA report discusses is polygenic: how you metabolise caffeine, how your body handles folate, how your variants compare with those described in published population research.

The distinction is not a consolation prize. Monogenic disease requires changing the sequence. Polygenic risk is modified by nearly everything downstream of it — which is the entire premise of nutrigenomics and lifestyle genetics.

This article is general educational information about a field of research. It is not medical advice, not a recommendation for or against any treatment, and not promotional material for any medicine. Treatment decisions belong with a qualified clinician.

DeepDNA interprets the genome you already have. It does not edit anything, and it is not a diagnostic service or a substitute for clinical genetic testing. If you want to understand what your existing data says, see a sample report. If you need a clinical answer, you need a clinician.

Frequently Asked Questions

How does CRISPR work in simple terms?

A guide RNA — a short, programmable sequence — leads the Cas9 protein to a matching stretch of DNA, and Cas9 cuts both strands there. The cell then repairs the cut, and that repair is what actually changes the gene: patched sloppily, the gene stops working; supplied with a template, it can be rewritten. Newer methods such as base and prime editing skip the cut entirely and alter the sequence more directly.

What diseases can CRISPR treat?

As of August 2026, one CRISPR therapy is approved anywhere in the world, covering sickle cell disease and transfusion-dependent beta-thalassemia. Clinical trials are underway for hereditary angioedema, ATTR amyloidosis, high cholesterol, and blood cancers via edited immune cells. All are conditions caused by a single gene or targeting a single well-defined molecular pathway.

Can CRISPR fix my genes?

Not as a consumer service. CRISPR is available today either as an approved therapy for specific severe conditions, or through clinical trials with strict eligibility criteria. There is no route by which an individual can request an edit to a variant found in a DNA report, and consumer genotyping is not diagnostic in the first place.

What is the difference between genetic testing and gene editing?

Genetic testing reads your DNA and reports what it says. Gene editing changes the sequence. They are separate technologies: reading your genome does not create a pathway to editing it, and a consumer genotyping array does not identify the kind of mutations that current CRISPR therapies correct.

Does CRISPR change my children's DNA?

No. All approved and trialled CRISPR therapies are somatic, meaning they alter cells in the treated person's body only, and those changes are not inherited. Germline editing — modifying embryos, eggs or sperm — would be heritable, and Article 13 of the Oviedo Convention prohibits it in Europe, as does the EU Clinical Trials Regulation.

Is CRISPR legal in Europe?

Somatic gene editing is legal and regulated in Europe as an advanced therapy medicinal product, assessed centrally by the EMA. Casgevy received conditional European authorisation on 9 February 2024. Heritable germline editing is prohibited. Trials additionally have to satisfy EU GMO legislation, which is transposed differently in each member state and is a documented source of delay.

How much does CRISPR treatment cost?

The approved therapy has a US list price of $2.2 million and a UK list price of £1.65 million, with confidential discounts negotiated nationally. NHS England agreed reimbursement for beta-thalassemia in August 2024 and sickle cell disease in January 2025, Italy in September 2025, and German statutory insurers in May 2026. Cost is not the only barrier: manufacturing capacity limited treatment to roughly 60 patients worldwide as of February 2026.

Why can't CRISPR cure heart disease or diabetes?

Because those conditions are polygenic. CRISPR corrects a specific mutation at a specific location, which works when a single gene causes the disease. Common conditions arise from thousands of variants each contributing a small amount of risk, interacting with environment and behaviour. There is no single edit that addresses them.

Is CRISPR safe?

CRISPR therapy is approved and regulated, but it is not risk-free. In ex vivo treatments the dominant documented risk is the chemotherapy conditioning required before the edited cells are returned, rather than the editing itself. At the molecular level, large deletions at the target site, a p53 damage response, and off-target structural changes invisible to prediction algorithms are all documented and actively researched. Two patients have died in gene editing trials, in neither case attributed to the edit.

What is the difference between CRISPR, base editing and prime editing?

Standard CRISPR-Cas9 cuts both strands of DNA and relies on the cell's repair machinery, which reliably breaks a gene but corrects one only inefficiently. Base editing chemically converts one DNA letter into another without cutting both strands, handling a subset of possible changes. Prime editing writes a specified new sequence directly, using a guide RNA that encodes the desired edit. Both newer methods avoid the double-strand break that causes most collateral damage.

This article was created with AI assistance and reviewed by the DeepDNA editorial team.