This is a very interesting question and I had not thought about it before. I am reiterating the background of your question.
Cas9 cuts the both the strands of the target DNA at the same location thereby making a blunt end cut. Usually sticky ends and microhomologous ends lead to indels because of exonucleolytic clipping and addition of extra bases by pol-µ. The question is how do indels arise during NHEJ mediated re-ligation of blunt ends.
NHEJ may not always lead to indels. Most blunt end cuts can be repaired without error. Some non blunt cuts are also perfectly re-ligated; the best example for this is VDJ recombination. The binding of Ku proteins is essential for a perfect re-ligation (canonical NHEJ/C-NHEJ).
However, there are some variant non-canonical pathways which include alternative-NHEJ, backup-NHEJ and microhomology mediated end joining (MMEJ) (Bétermier et al. 2014). Some studies also refer to all of these pathways as alternative-NHEJ or altNHEJ. These alternative pathways involve resection of nucleotides by exonucleases such as Mre11 to produce overhangs. Imperfect alignment of these overhangs prior to ligation leads to indels. Sometimes non-templated addition of nucleotide happens by the action of terminal deoxyribonuclotide transferase (TdT). Pol-µ can also add random nucleotides at the 3' end (Lieber 2010).
In general, the propensity of DSBR via alternative NHEJ pathways would be higher when Ku activity is lower. Even with a moderate Ku activity, there is always some likelihood of alternative pathways in action. Moreover, it is to be noted that sticky end ligations are more efficient than blunt end ligations.
Reproduced from Bétermier et al. 2014