Why are you talking about markdown? LLMs output bytes and llama.cpp has to process those bytes correctly. What if the LLM outputs a null byte and causes a buffe

Why are you talking about markdown? LLMs output bytes and llama.cpp has to process those bytes correctly. What if the LLM outputs a null byte and causes a buffer overflow somewhere and starts writing into executable memory? Again, we are not talking about agents or your Python API script but instead talking about exploitable flaws within the inference engine itself. It wouldn't output `rm -rf`. It would output literal CPU instructions that llama.cpp would start executing directly. The payload would never get back to your Python script.