Genetic engineering in humans carries risks at every level, from the molecular to the generational. Some are technical problems that scientists are actively working to reduce, like the editing tool cutting DNA in the wrong place. Others are deeper and harder to solve, like the possibility that changes to embryos could ripple through future generations in ways no one can predict. Here’s what those risks actually look like.
Cuts in the Wrong Place
The most widely used gene editing tool, CRISPR-Cas9, works by guiding a molecular “scissors” to a specific spot in your DNA. The problem is that the guide isn’t perfectly precise. A single guide RNA can recognize DNA sequences that differ by as many as 3 to 5 base pairs from the intended target, which means there could be thousands of possible binding sites for a given guide in the human genome. The system essentially has more energy than it needs to find its target, so it latches onto sequences that look similar enough.
In laboratory studies on human cells, the numbers can be striking. When researchers tested standard CRISPR on human cell lines, off-target mutation rates at some sites ranged from 9.4% to 93.6%. In one experiment using guides with a single mismatch, the tool still edited the wrong location roughly 50 to 60% of the time. These unintended cuts can cause deletions, inversions, and translocations, where chunks of chromosomes break off and reattach in the wrong place or on the wrong chromosome entirely.
Newer versions of the system are bringing those numbers down significantly. Modified forms of the Cas9 protein have reduced off-target editing to under 4% at most tested sites. But “reduced” is not “eliminated,” and in a therapeutic context where billions of cells are being edited, even a low percentage translates to a large absolute number of unintended mutations.
Damage at the Target Site
Even when CRISPR cuts exactly where it’s supposed to, the repair process itself can go wrong. After the tool makes a double-strand break in the DNA, the cell tries to fix it. The most common repair pathway is error-prone, frequently introducing small insertions or deletions at the cut site. More concerning, researchers have documented large-scale genomic deletions at intended target sites, where the cell loses substantial stretches of DNA during repair. These on-target errors are harder to screen for than off-target cuts because they happen at the “correct” location, making them easy to miss with standard testing.
The Mosaicism Problem in Embryos
When gene editing is applied to embryos, the timing of the edit matters enormously. If the CRISPR machinery doesn’t finish its work before the single-cell embryo starts dividing, different cells end up with different genetic edits. This is called mosaicism, and it’s far more common than most people realize.
In embryo editing experiments (conducted in cattle, which serve as a close model), mosaicism rates were 94 to 100% depending on the form of CRISPR used. Researchers found that editing continued in some subset of cells even after the first cell division, producing embryos with three or more distinct genetic versions mixed together. In practical terms, this means one organ might carry the intended correction while another carries an unintended mutation, or no edit at all. For any future attempt at human embryo editing, mosaicism remains one of the most stubborn technical barriers.
A Possible Link to Cancer
One of the more unsettling findings in recent years involves a well-known tumor suppressor gene called p53. This gene acts as a safety switch: when a cell’s DNA is damaged, p53 triggers the cell to either repair itself or self-destruct. CRISPR’s double-strand breaks activate this same alarm system. That’s normally a good thing, but it creates a perverse selection pressure.
Cells with functioning p53 respond to CRISPR-induced damage by slowing down or dying. Cells with broken p53, which are already one step closer to becoming cancerous, survive and keep dividing. Research published in Cancer Research demonstrated that applying CRISPR to a mixed population of normal and p53-deficient cells enriches the population for the defective cells. The stronger the DNA damage response triggered by the editing, the greater the enrichment. In other words, the very act of editing could give pre-cancerous cells a competitive advantage over healthy ones.
This is partly why the FDA has required long-term cancer monitoring for approved gene therapies. For the first CRISPR-based therapy (Casgevy, approved for sickle cell disease), regulators mandated postmarketing studies specifically to assess the long-term risk of blood cancers related to the editing process.
Immune Reactions to the Delivery System
Getting the editing machinery into human cells requires a delivery vehicle, and the most common ones are modified viruses. These viral vectors can trigger immune responses that range from mild to dangerous. Over 90% of humans already carry antibodies against some of the viral types used in gene therapy. Some of those antibodies are neutralizing, meaning they can destroy the therapy before it reaches its target cells, potentially rendering an expensive treatment useless on the very first dose.
Beyond efficacy loss, viral vectors can activate broader immune responses. The immune system may attack cells that have taken up the vector, causing inflammation or, in severe cases, triggering a systemic inflammatory reaction. The FDA identifies immunogenicity as a major problem across gene therapies, noting that it can cause diverse immune-related toxicities through multiple branches of the immune system.
The Body Can Silence New Genes
Even when a new gene is successfully inserted and begins working, the body may shut it down over time. This phenomenon, called gene silencing, occurs when the cell’s own defense mechanisms recognize the inserted DNA as foreign and chemically inactivate it. In one experiment, cells engineered to produce a therapeutic enzyme showed a more than 1,000-fold drop in production within 30 days of being transplanted into living tissue.
Silencing is especially common with certain types of viral delivery vectors, because the cell has evolved to recognize and suppress viral-like genetic sequences. Long-term expression lasting years rather than weeks has been difficult to demonstrate for some vector types. The location where the new gene lands in the genome also matters: some insertion sites are prone to silencing while others are not, and controlling this precisely remains a challenge. For patients, this means a gene therapy that works initially could fade over months or years, potentially requiring repeated treatment or leaving the underlying condition unaddressed.
Side Effects From Current Therapies
The real-world safety profile of approved gene-editing therapies gives a concrete picture of what patients actually experience. In the clinical trials for Casgevy, which involved 44 patients, the side effects were largely tied to the preparation process rather than the gene edit itself. Before receiving their edited cells back, patients undergo a harsh chemotherapy regimen to clear out their existing bone marrow and make room for the corrected cells.
Nearly all patients (98%) experienced mouth sores. About 70% had nausea, 66% had musculoskeletal pain, and 55% developed febrile neutropenia, a fever caused by dangerously low white blood cell counts. Serious complications included pneumonia in about 9% of patients, and notably, delayed recovery of platelet counts, which affects blood clotting. Three patients experienced sickle cell crises even after treatment. These side effects reflect the intensity of the overall procedure, not just the gene editing component, but they’re inseparable from the therapy as it currently exists.
Heritable Changes and Future Generations
The risks above apply to somatic editing, where changes are made to a patient’s body cells and cannot be inherited. Germline editing, which would alter eggs, sperm, or embryos, introduces an entirely different category of risk: any mistake becomes permanent in the human gene pool.
An off-target mutation in a somatic cell affects one person. The same mutation in a germline cell could be passed to children, grandchildren, and beyond. Because gene frequency dynamics in populations are extraordinarily complex, the downstream effects of introducing modified genes are essentially impossible to model. A gene that appears harmless or beneficial in one generation could interact with other genetic or environmental factors in ways that only become apparent decades later. Conversely, genes that seem worth eliminating may carry hidden protective functions that aren’t yet understood.
This unpredictability is why germline editing occupies a unique place in risk discussions. The effects are not just difficult to reverse; for all practical purposes, they are irreversible once a modified gene enters a breeding population. The World Health Organization established a global expert committee in 2018 specifically to develop governance frameworks for this issue, and no country has approved heritable human genome editing for clinical use.
Why Long-Term Monitoring Matters
Many of these risks play out over years or decades, not days. The FDA requires patients who receive gene therapies to be monitored for up to 15 years after treatment. For the first five years, this means annual in-person examinations with blood testing for persistent viral vector sequences and screening for new cancers, neurological disorders, autoimmune conditions, and blood disorders. For the following ten years, patients are contacted at least annually by phone or questionnaire.
This 15-year window applies to therapies using integrating viral vectors, genome editing products, and several other categories. It reflects a candid acknowledgment that the full consequences of genetic engineering in humans simply aren’t knowable on the timescale of a typical clinical trial. The monitoring isn’t just precautionary bookkeeping. It’s the primary mechanism for catching problems like delayed-onset cancers or gradual loss of therapeutic effect before they become widespread.