CRISPR-Cas9 turned the genome into an editable document. What's changed since is who's doing the editing: AI is no longer just picking targets for human-designed enzymes — it's designing the enzymes themselves, predicting their molecular contacts, and steering next-generation editors that fix DNA without ever cutting it.
For most of CRISPR's first decade, AI's role was modest: pick the best guide RNA, predict off-target risk, rank candidate edits. That's changed. Researchers have used AlphaFold3's predictions of the molecular contacts a genome-editing enzyme makes with DNA to directly improve that enzyme's targeting selectivity — using a structure-prediction model built for proteins to make a better gene-editing tool.
Further out, protein language models (the same family of model behind AlphaFold) are now running "in-silico directed evolution" — proposing thousands of variants of a CRISPR enzyme computationally and ranking them for activity before a single one is synthesized in a lab. That process has already been used to design functional CRISPR enzymes with no natural counterpart: proteins that never existed until a model proposed them. Separate work applies explainable AI models directly to guide-RNA design, aiming to flag off-target risk before synthesis rather than after.
AI is no longer just picking targets for human-designed enzymes — it's designing the enzymes themselves. — AI-ification.AI
Classic CRISPR-Cas9 works by cutting both strands of DNA and letting the cell's own repair machinery patch the break — effective, but imprecise and sometimes disruptive. Base editors skip the cut entirely, chemically converting one DNA letter to another directly. Prime editors go further: a "search and replace" system capable of precise substitutions, insertions, and deletions without a double-strand break or a donor DNA template at all.
Delivery remains the hard engineering problem. Split prime editors delivered via dual AAV vectors have hit editing frequencies of 40–50% in multiple mouse tissues, and engineered virus-like particles have successfully corrected a pathogenic mutation in a mouse model of inherited blindness — both real, published results, not projections.
Beam Therapeutics has reached FDA alignment on an accelerated approval pathway for a base-editing treatment for hereditary ATTR amyloidosis, delivered via lipid nanoparticle. The first patient was dosed in June 2024, and more than 25 participants have been treated since — base editing's move from mouse studies into a real accelerated-approval track. It follows Casgevy, the CRISPR-Cas9 therapy for sickle cell disease (which targets the same HBB gene featured in the live genome browser on the homepage) that became the first CRISPR-based cure ever approved, in 2023.