If that's the case then you were doing it wrong. False positives from Bloom filters are always possible, by design.
Bloom filters are not about depending on true positives, but rather depending on true negatives.
The proximate cause might have been a false positive, but I think it's more fruitful to think of the likely root cause: picking a data structure without thinking through/understanding the ramifications ultimately cost you 100k USD.
Are you allowed to explain why exactly that happened?
Used a bloom filter to see if a password was correct to get into their bitcoin stash.