A Science paper reports that AI-designed, synthetic RNA-guided nucleases can achieve activity that matches or outperforms natural reference enzymes, broadening the CRISPR toolbox beyond naturally evolved architectures. Using an approach combining an inverse protein-folding model with evolution-informed constraints, researchers generated novel variants for a TnpB family of RNA-guided nuclease proteins. The work demonstrates functional activity in cells while emphasizing how structure-guided protein design can sample more divergent sequence space than sequence-only modeling. By expanding the set of feasible nuclease properties, the method aims to reduce engineering constraints encountered when multi-domain nucleases need improved specificity and compatibility. For gene-editing developers, the advance adds another route to diversify editing enzymes—potentially relevant for minimizing off-target effects and tailoring tool behavior to specific delivery or therapeutic contexts.