CRISPR-Cas9 gene editing directs a bacterial enzyme, Cas9, to cut DNA at a location specified by a short guide RNA sequence that binds through Watson-Crick base pairing to a matching stretch of the genome. Because the guide RNA can be redesigned to target virtually any twenty-nucleotide sequence, the system in principle allows a single defective gene to be corrected without disturbing the rest of the genome. In practice, however, Cas9 will also cut at off-target sites where the genomic sequence differs from the intended target by one or even several nucleotides, provided the mismatches fall in less critical positions of the guide-target pairing. The frequency of off-target cutting is not fixed; it depends on an interaction between the guide sequence's own properties and how accessible the candidate off-target site is within the cell's chromatin. A guide RNA with a mismatch near the enzyme's cutting site is tolerated far less often than one with a mismatch near the opposite end, because pairing stability near the cut site matters disproportionately to whether the enzyme completes its cut. Compounding this, a genomic region that is tightly packaged and transcriptionally silent is cut less readily than one that sits in open, actively transcribed chromatin, even when the two regions carry an identical number of mismatches to the guide. Consequently, predicting off-target risk from sequence similarity alone, without accounting for where a candidate site falls in the genome's three-dimensional packaging, systematically understates the risk in open chromatin and overstates it in condensed chromatin, an asymmetry that has forced editing protocols to combine sequence-based prediction with chromatin accessibility mapping before a guide RNA is judged safe for therapeutic use.