For me, a moral agent is not one that can 'feel responsible', but one that I can affect, by changing my own behavior, when it does 'something bad'.
The discussion of internal states is less useful. Also, the discussion of morality of people I can't affect is less useful.
If a person lobs avalanche prevention ordinance upon an unstable mountain slope thereby preventing the demise of the hikers I invented earlier, that person's actions might be said to have affected the behavior of the mountain.
If the person lobs ordinance after the invented hikers have met their fate, that person may well change the mountain slope geometry in such a manner that unique unstable accumulations are no longer physically possible [while others may be].
Clearly neither seems like a case where there is moral agency in the mountain. We get moral agency when the person lobbing ordinance onto the mountain slope is doing or forgoing the activity in the presence of hikers.
Such a scenario might well involve individuals engaged in banal evil and because individuals are moral agents, that's where blame and praise may accrue.
Anthropomorphizing a tobacco company muddies our thinking. A trade agreement is a document. The consequences of following the ideas in that document may be detrimental. But it is not a moral agent.