It’s not really clear to me from reading the article what it did here, but it sounds like the judge asked the bot if there is precedent. If that’s the case, what if the issue is nuanced and the precedents are conflicting - the bot might provide favourable rulings which check out under review, but omit unfavourable ones.