This isn’t about tricking people into using the wrong model. Training data can poison a model in ways that cannot be reasonably detected or trained out generica

This isn’t about tricking people into using the wrong model. Training data can poison a model in ways that cannot be reasonably detected or trained out generically. So even if you are using the right model, it could be poisoned without anybody knowing.