Genuinely curious of your thoughts!
Posting an order is a statement of intent. If you allow a machine to post those in your name you take responsibility for the claims made by that machine. Discovery of that "one magic trick" by ML reminds me of the way toddlers learn all kinds of mischievous "life hacks" like "I can reach goal X by dropping object Y" before they start to respect more cooperative forms of interaction. I am skeptical of allowing toddlers on the trade floor. And if you did allow then, you would want to have mechanisms to make their parents take responsibility while their children are not yet able to.
If the decision-makers at the exchanges running the show were not so much closer with those trading for trade than with those trading for actual ownership, they would have curbed this abuse very early. Maybe by introducing a sufficiently low upper limit to the volume of offers that can be cancelled (relative to the volume of offers that are followed through), or some form of progressive cancellation fee that would protect the market from this form of abuse. The observation that only external supervision put an end to it (instead of the "house rules" of the exchanges) makes it difficult for me to dismiss as paranoid the claims made in the discussion here that he just lacked the right friends to pull this off.
I think it's safe to assume that this is not a "one daring bet" kind of manipulation, like e.g. badly disguised insider trading could be, it is rather wealth by a thousand papercuts. The pattern is very unlikely to be worthwhile without excessive repetition and there are only so many million times where you can believable claim that you wanted, then you didn't, and than you wanted the opposite, all in carefully timed lockstep.