On Safe Sequential Decision-Making In Adversarial Environments