What does the current evidence actually show about long-term safety?
The short answer is: the evidence shows promising short-term safety, but it is far too early to declare base editing risk-free. In a systematic review of over 170 patients treated with gene editing (including base editing) for sickle cell disease and beta-thalassemia, no therapy-related malignancies or confirmed harmful off-target edits were reported, but the median follow-up was only about 18 months, with the longest follow-up just over 4 years [1]. That means we have good data on what happens in the first year and a half, but almost no data on what happens after 5 or 10 years. Another systematic review of 148 infused patients across multiple gene therapy platforms (including base editing) reached the same conclusion: long-term safety remains an unresolved barrier [3].
The situation is similar for in vivo (directly injected) base editing. In a nonhuman primate study of VERVE-101, a base editing therapy for high cholesterol, the therapy was well-tolerated and produced durable effects for up to 476 days (about 1.3 years), but again, follow-up beyond that point is not yet available [2]. So while the early signal is encouraging, the window of observation is simply too narrow to rule out late-emerging problems.
Which specific long-term risks are hardest to detect in current trials?
Two categories of risk are particularly difficult to assess with short follow-up: off-target edits that could trigger cancer years later, and unintended 'bystander' edits near the target site. One study using lipid nanoparticles to deliver adenine base editors to blood stem cells found that while one guide RNA showed minimal off-target activity, another guide had broader off-target effects, including editing at exonic (gene-coding) sites, though none were near known cancer genes [4]. The same study noted that bystander edits—unintended changes to nearby DNA letters—are a known limitation of the base editor variant they used, though they occurred in non-coding regions and were judged low-risk [4]. The problem is that proving a low risk of cancer requires following thousands of patients for many years, because some cancers take decades to develop. Current trials, with their small numbers and short follow-up, simply cannot provide that proof.
Another risk is immune response to the editing machinery itself. A 2026 review of CRISPR-based therapies notes that immune responses to Cas proteins (the 'scissors' used in editing) remain a challenge, and these could potentially cause inflammation or rejection of edited cells over time [5]. This is not something that would show up in a 6-month safety report, but could become a problem with repeated dosing or in patients with pre-existing immunity.
How do the trial designs themselves limit what we can conclude about long-term risk?
The current trials are almost all single-arm studies (no comparison group receiving a placebo or standard treatment), which makes it difficult to attribute rare adverse events to the therapy versus the underlying disease or the conditioning chemotherapy used before cell infusion [1][3]. For example, in the sickle cell and beta-thalassemia trials, one death and all serious adverse events were attributed to the busulfan conditioning (chemotherapy given before the edited cells are infused), not to the editing itself [1]. But without a control group, it is hard to know whether the editing might be contributing in subtle ways. The same review notes that delivery complexity—including the need for chemotherapy, apheresis (blood cell collection), and manufacturing—remains a critical barrier [1]. These procedural risks are separate from the editing risk, but they complicate the overall risk assessment.
Furthermore, the patient numbers are small. The largest systematic review here included just over 170 treated patients across all studies [1], and another included 148 infused patients [3]. Detecting a rare side effect that occurs in, say, 1 in 500 patients would require a much larger population. So the absence of evidence of harm is not yet evidence of absence of harm.
About These Sources
This answer is built on 5 peer-reviewed studies — published from 2023 to 2026, 4 from 2024 or later — selected as the most relevant from 5 studies that passed quality screening, drawn from 93 papers retrieved from a database of over 500 million.
Sources used in this answer
Gene Editing Therapies for Sickle Cell Disease and β-Thalassemia: A Systematic Review of Clinical Outcomes and Safety
In a systematic review of over 170 patients treated with gene editing (including base editing) for sickle cell disease and beta-thalassemia, no therapy-related malignancies or confirmed harmful off-target edits were reported, but median follow-up was only ~18 months (longest >4 years), and all serious adverse events were attributed to busulfan conditioning, not editing.
Efficacy and Safety of an Investigational Single-Course CRISPR Base-Editing Therapy Targeting
In a nonhuman primate study of VERVE-101 (an in vivo base editing therapy for high cholesterol), the therapy was well-tolerated and produced durable effects up to 476 days, but follow-up beyond that point is not yet available, and germline editing was assessed in animal models.
Efficacy, Safety, and Treatment-Delivery Feasibility of Autologous Gene Therapy for Sickle Cell Disease: A Systematic Review With Descriptive Synthesis of Clinical Trials.
A systematic review of 148 infused patients across multiple gene therapy platforms for sickle cell disease (including base editing) concluded that long-term safety, fertility, organ outcomes, and global scalability remain unresolved implementation barriers, with certainty limited by single-arm designs, small samples, and incomplete long-term follow-up.
Toward ‘n of one’ gene therapies: Scalable ex vivo adenine base editing via lipid nanoparticles in a sickle cell disease model
In a study using lipid nanoparticles to deliver adenine base editors to human blood stem cells, one guide RNA showed minimal off-target activity while another had broader off-target effects including exonic edits (none near known oncogenes); bystander edits were observed but in non-coding regions, judged low-risk.
CRISPR–Cas Systems in Human Disease Therapy: Advances, Clinical Applications, Limitations, and Future Directions
A 2026 review of CRISPR-based therapies notes that clinical translation is constrained by off-target activity, delivery inefficiency, immune responses to Cas proteins, editing heterogeneity, and uncertainties regarding long-term safety.