Researchers reported in Science a new AI-designed strategy for building functional CRISPR-like nucleases that can rival or outperform natural enzymes. The work describes structure- and evolution-guided design of synthetic, RNA-guided nucleases where activity is preserved while sequences can diverge substantially from reference proteins. The approach is positioned as a way to expand the “CRISPR toolbox” beyond enzymes found in nature. As a test case, the team designed variants for TnpB, a family of CRISPR-Cas12-like nucleases, using an ESM Inverse Folding model (inverse protein-folding) combined with evolution-informed residue constraints to maintain compatibility for guide RNA and DNA recognition. Candidate proteins were then screened for editing activity in cell and nonclinical systems, including plant and animal cells. Separate reporting highlighted a platform from Nobel laureate Jennifer Doudna’s team in Science that generated non-natural TnpB enzymes capable of cutting DNA in multiple biological contexts. Together, the studies point to a workflow where AI proposes enzyme sequences around functional constraints, and lab assays validate whether those designs maintain programmable activity. For biotech, the immediate significance is methodological: if AI-guided nucleases can be reliably engineered with lower off-target behavior or new biochemical properties, the same design loop could be used to tailor editing tools for different therapeutic delivery constraints and safety profiles.