Two previous papers (Wu et al and Kuscu et al) showed that Cas9/guide RNA complexes are normally resident on many genomic sites, the vast majority of which do not get cleaved by Cas9. Thus, binding does not automatically lead to cleavage - far from it. The data in this paper clearly indicates the requirement of the PAM-proximal "seed" portion of the protospacer for initial Cas9/sgRNA binding, while the distal protospacer does not need to match to permit binding. However, the additional pairing of the more distal portion of the protospacer is required for target cleavage. Thus binding and cleavage are distinct steps.
One important implication of these studies is in regard to using cleavage-deficient Cas9, e.g. with both DNAse domains mutated, to physically localize fusion complexes (GFP, transactivation domains, etc) to genomic targets for reasons other than genomic editing. Examples include fluorescent labeling of telomeres and upregulation of specific target genes. (Cheng et al and Perez-Pinera et al).
The results from this and the previous papers now clearly suggests that for CRISPR-Cas9 applications that only require binding to the target, not cleavage, off-target effects may be more widespread. Whether that will be an serious problem, and maybe it won't be, will probably be very dependent on the application in question.
In any event these papers provide data that may point the way to further increasing CRISPR/Cas9 specificity, in a variety of different applications and settings.