Since the earliest days of genome editing, scientists have wrestled with a stubborn paradox: the more powerful the molecular scalpel, the less reliably it cuts only where intended. A team publishing in Nature has now used artificial intelligence to dissolve that trade-off, developing a framework called ContactSeek that pinpoints the precise amino acids governing whether a genome editor binds to the right DNA sequence or wanders into unintended territory. By changing as little as a single molecular letter in tools like Cas9 and Cas12a, they achieved reductions in off-target editing of up to 99