I think GLM 5.2 might have found this. See Poelstra's response in this github thread: https://github.com/ElementsProject/elements/issues/1325 . What if someone created a transaction with an OP_RETURN that has a pedersen commitment that has some big negative value? If it gets included in the balance equation, that can work *only if* the range proof is not attached to that output.

See https://blockstream.info/liquid/tx/f24a4b179b5cc7e88b25a763911f7cbdf2bf45d1d1b5ab611e94461cef0a183f?expand - this is the transaction at which people are saying there's a consensus split.