This assumption is usually made in game theoretic versions of contract theory. If utility is unbounded from below agents play a game repeatedly then there is some path in which behavior can be enforced so that incentive compatibility holds. I may have phrased this incorrectly - it's been a while since I've looked at game theory.
Basically, all it says is that we can enforce good behavior with sufficient threats. On the flip side, one would also think that if faced with the possibility of dying, we would make a decision to preserve our lives. In equilibrium, with perfect knowledge and perfect rationality then workplace safety would not be an issue. Of course, we don't have perfect knowledge or perfect rationality so regulation is required. But can regulation perform better than a worker's private knowledge of workplace issues?
This post is basically in response to this NYT article on workers' concerns aboard the Deepwater Horizon before it blew up:
A confidential survey of workers on the Deepwater Horizon in the weeks before the oil rig exploded showed that many of them were concerned about safety practices and feared reprisals if they reported mistakes or other problems.
In the survey, commissioned by the rig’s owner, Transocean, workers said that company plans were not carried out properly and that they “often saw unsafe behaviors on the rig.”
Some workers also voiced concerns about poor equipment reliability, “which they believed was as a result of drilling priorities taking precedence over planned maintenance,” according to the survey, one of two Transocean reports obtained by The New York Times.