I wonder what strategies need to be employed to feel confident that you have even surfaced all the problems. Presumably there is a list of unmitigatable problems or worst defended attack vectors. But that is probably the hidden success of our last century, that and tight machine tolerances it's everywhere.
Well, to some extent it is a simple calculation. Depending on the amount and radioactivity of an isotope, you get a certain amount of heating.
If you go past some amount, you start requiring active cooling to stop it from increasing faster and faster in temperature and reacting more violently - this may happen in traditional power plants. If instead you limit the reactor to a much smaller amount of fuel, it starts losing more heat to air than it produces from radioactivity, and this problem goes away.
Of course, the problem is then to obtain that heating while the reactor is operational, but I understand the solutions are known.
This is similar to why fusion reactors are not a risk of becoming nuclear weapons: the system requires external power input just to keep the reaction going. So, in the event of a problem, you'll lose the reaction, instead of it running away uncontrollably.
Of course, there are other risks. For example, a fusion plant may suffer a breach of the reaction chamber, and all that energy will violently explode, releasing radioactive tritium and bits of radioactivated materials from the reactor structure around. I would guess MMRs have similar bad-but-not-catastophic failure modes.
