Is that a big risk though? I imagine that for a patient (and with that a specific genome with specific editing errors) you could extract some cells, grow an in vitro culture, and test the CRISPR on that? You could then sequence the altered genome and see how much genes were "collateraly edited" and only if the result is non-risky apply the medication to the patient.

T-Cells are remarkable tricky to grow in-vitro. Most of them only grow after presented the particular antigen (for the non-biologists: the part of the virus/bacteria which they recognise) which they are keyed to. I haven't looked at the actual proposals, but I figure they'll take T-Cells, use CRISPR on those directly, clean them off (so you don't accidentally CRISPRize other cells in the body) and re-inject them directly so they won't die off.

You could sequence part of them before insertion, if you manage not to kill the rest of the T-Cells with the wildly differing in vitro environment, but by that point you're playing a numbers game. The non-specific CRISPR action (called off-target mutagenesis in the literature) is a low probability event, but if you sequence part of the T-Cells, you could easily miss one with a bad (read: non-lethal) off-target mutation.

That makes sense! Thanks for the insight!