That’s true, but there’s a difference between people using an exploit that is not natively supported vs using op_return which is a feature with the purpose to allow arbitrary data. By increasing the op_return limit you are now signalling that relaying large blobs of continuous data is a feature supported by the network. In contrast, shoving split up chunks of data into the witness section where it is not designed to be is not an intended use of the network. The argument is that: now relaying large continuous chunks of data is natively supported, any legal/regulatory plausible deniability is gone. From a technical standpoint this change also makes it easier for bad actors / criminals as they don’t have to split their files into multiple chunks.