But the issue of violating the Wikipedia policy doesn't factor much into the calculation.

The fact that the issue violates Wikipedia policy is an essential part of why doing as you propose would be likely to have a negative impact on MIRI's reputation.

(For the avoidance of doubt, I don't think this is the only reason not to do it. If you use something that has policies, you should generally follow those policies unless they're very unreasonable. But since ChristianKI is arguing that an expected-utility calculation produces results that swamp that (by tweaking the probability of a good/bad singularity) I think it's important to note that expected utility maximizing doesn't by any means obviously produce the conclusions he's arguing for.)