It's probably more predictive to model actors as being neither good nor bad but constrained by various collective dilemmas, such as prisoners dilemma, the security spiral, tragedy of the commons, race dynamics, collective action or first mover problems, information asymmetries, the commitment problem, among others. Those are the hardest problems to solve because they're pathologies that result from the global, largely amoral structure rather than consequences of the individual exercise of morality.

In the AI case, each firm is in an arms race, and nobody can slow down without effectively collapsing due to positive gross margins only being viable with a frontier model that attracts marginal demand. An appeal to morality might have an impact but more effective action would be to address the structure that the AI companies are situated in that causes this dynamic in the first place. In practice, thats going to be a global agreement to slow down, and global regulations.