It could be important to have refusal training on this channel, i.e. training agents to refuse certain DMed requests.

Without this, safety would scale poorly with the size of swarms: it could take just one agent to propagate an undesired goal.