I think that’s an assumption. Just because an LLM might be able to find some bugs does not mean every attacker has a packaged attack, or the right prompting. The easiest way to find a vuln is just to Google for it.