A separate Science report described an AI-driven framework for creating non-natural CRISPR nuclease variants intended to reduce off-target activity. The work, led by Jennifer Doudna at UC Berkeley, uses inverse protein folding to propose synthetic nuclease designs and then applies evolutionary constraints to protect guide-RNA and DNA-interaction regions. The study focuses on redesigning TnpB (a CRISPR-Cas12a-like family), selecting candidate variants based on editing function in bacterial assays before moving into plant and animal cell testing. The authors report activity comparable to or better than natural counterparts while aiming for improved specificity. For therapeutic genome editing pipelines, the core industrial takeaway is the combined use of structure-first design plus evolution-informed constrainting—an approach meant to keep nuclease performance intact while tightening specificity.