Just off the top of my non-expert head about the security of LLM:1. Security holes and malicious elements in the models'...

Just off the top of my non-expert head about the security of LLM:1. Security holes and malicious elements in the models' corpus.2. Holes, bugs and vagueness in their interpretation of natural language prompts3. Their inability to separate commands from data.4. Security holes in the systems that interface the LLM engine with users.5. Complex interactions between all of the above leave fertile ground for systemic security holes and bugs that may not be findable or fixable.#LLM #infosec