Comment by gsquaredxc

I have a hard time viewing prompt injection as malware. LLMs are unpredictable and there are many different prompts that can unintentionally cause unexpected behavior. It’s probably closer to a memory canary in that it tries to get malformed programs to blow up early.