The rules of evidence place a lot (honestly an unreasonable amount) of weight on the value of eyewitness evidence, and contemporaneous reports made from the same. The question for the courts will be, does an AI summary of a video recording have the same value as a human-written report from memory?
I agree that this is good use of AI, but would suggest that th courts should require an AI report to basically have the body cam recording stapled to it, ideally with timestamped references in the report. AI transcriptions are decent, but not perfect, and in cases where there could be confusion the way the courts treat these reports should allow for both parties to review and offer their own interpretations.