Comment by rcxdude

Yes. The point with this attack is that you can't really avoid it by 'aligning' the LLM any better, nor really by filtering the actions it takes better.