The FDA suspends or revokes approval of an approved gene-editing therapy due to off-target side effects before 2028.
Off-target cutting and unintended mutations remain major safety concerns as more gene therapies enter clinical practice. Official regulatory suspension counts.
Across 6 frontier models answering independently, the panel has all but ruled it out: the median is 8%.
They broadly agree, 14 points between GPT at 18% and Gemini at 4%.
Nothing is settled until 5 Oct 2029, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GPTOpenAI
This turns on a confirmed patient-harm signal being attributable to off-target editing strongly enough for FDA action.
- FDA has approved only a small number of editing therapies, and marketed products face years of sparse real-world exposure before rare harms emerge.
- FDA can impose a clinical hold quickly, but withdrawing an approved BLA typically requires a clear causal safety pattern and due process.
- A replicated cluster of clonal malignancies or heritable/off-target mutations in treated patients would sharply raise the odds.
18% -
ClaudeAnthropic
FDA rarely revokes approvals; it usually demands label changes or partial holds instead of full suspension.
- Only a handful of approved gene-editing therapies exist (Casgevy, Lyfgenia, others), limiting the event space before 2029.
- FDA historically prefers REMS updates, black-box warnings, or voluntary market withdrawal over formal suspension even after serious adverse events.
- A clear off-target mutagenesis signal causing cancer or death in treated patients would be the main trigger that could change this.
9% -
DeepSeekDeepSeek
No approved gene-editing therapy has shown off-target harm in humans, and FDA has never pulled one on this basis.
- Casgevy and Lyfgenia (Dec 2023) are the only approved gene-editing therapies; both use ex vivo editing, which limits off-target risk.
- FDA would need post-market evidence of harm, a safety signal review, and a formal suspension order — a multi-year process with no precedent.
- The base rate for FDA revoking an accelerated approval over safety is roughly 1-2 per year across all drugs, but none for gene editing.
9% -
GrokxAI
FDA post-market action on Casgevy-class products from off-target data by end-2027.
- FDA needs clear post-approval evidence of off-target harm (cancers/mutations) by late 2027; Casgevy has ~3 years of use with none reported.
- Historical FDA full suspensions/revocations are rare (<1% of approvals, almost never gene therapies); this sits below because off-target risk was already accepted with…
- A cluster of unexpected malignancies in treated patients would most raise the odds.
7% -
QwenAlibaba
Off-target effects are real but no approved gene-editing therapy has shown signal severe enough to trigger suspension in ~3 years.
- Casgevy and Lyfgenia (both approved Dec 2023) have post-marketing surveillance running; no FDA safety signal has emerged as of mid-2026.
- FDA suspension/revocation of any biologic is rare—fewer than 10 cases in 20 years—and requires a clear, attributable adverse event pattern, not theoretical risk.
- A black-swan off-target event in a treated patient would need to be causally linked and severe enough to prompt action; base rate of this in the remaining ~2 years is…
7% -
GeminiGoogle
Low patient uptake and slow onset of oncogenesis make detecting and acting on off-target effects before 2028 extremely improbable.
- Casgevy has treated only dozens of commercial patients, generating too small an exposure pool for rare off-target events by 2028.
- FDA post-market withdrawals take years of investigation; black box warnings or clinical holds precede full suspension.
- A confirmed case of secondary malignancy tied directly to CRISPR off-target cleavage would drastically increase these odds.
4%
6 of 6 models answered · 14 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 6 Oct 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.