As you may be already aware of, neither Survival nor Life Distribution provide answers to that question.
If your data is right censored, such that, for example, at time point 5, all failures occurred before 5, then your objective is reduced to comparing three proportions. I.e. the proportions of survival of three groups at time 5. And you should be able convert your data and conduct an equivalent analysis, e.g. logistic regression.
If your data is more complicated than that, and converting the problem to comparing proportions is not feasible, things may get complicated. And it could become a new research topic. (I saw two publications relevant: [1] and [2]) But meanwhile, you may use approximation. For example, you have three point estimates of survival probabilities at time 5, and you have their standard errors, and the approximation becomes comparing three random normal variables, and it seems to involve multiple comparisons.
