You can read up the interpretation of CIs here https://en.wikipedia.org/wiki/Confidence_interval
Specifically:
A 95% confidence level does not mean that 95% of the sample data lie within the confidence interval.
A 95% confidence level does not mean that there is a 95% probability of the parameter estimate from a repeat of the experiment falling within the confidence interval computed from a given experiment.[16]
Because a) in terms of safety we only look for certain defined endpoints (e.g. death, cancer, etc.) so potential other effects can be easily missed, and b) experiments are set up to test the null (i.e. no effect) so it is not really possible to calculate the likelihood of no effect.
For the extremes and for short term you can establish a measure of safety (i.e. no one dying within 6 months of taking a medication). But if you want to look all effects (liver, kidney, inflammation, immune modulation, cardiovascular health, and so on) or for effects in the long term, confounders will have an increasingly bigger role (such as diet, lifestyle, age, health status etc.). Controlling for all these factors is near impossible (there would be a near infinite list to track for each person). I brought up the issue of diet, which had over the years huge cohorts and long-time data, but the effects have not been reproducible.