I'm sure Anthropic of all companies don't do that, since using mechinterp as a training target will make the the result uninterpretable.

They're definitely willing to consider it. Read the parts of the paper where they use J-space interpretations to train more ethical behavior into the model by interrupting it.