Do safety monitors survive fine-tuning?

When someone fine-tunes a model, does a safety monitor fitted to the original still read the new model correctly?

From the series: Align your AI Model