OP here: for those interested is not that an LLM found the bugs the fuzzer found them. My take on this mattter was to implement as much information theory algorithms to try to extract as much structure with statistical importance from the binary being fuzzed as possible, also port as much features from the other fuzzers as possible. I can not take honestly fully credit for this work since I made it with AI, but I has taken two months and 1100 commits and several hours of my time. My developing process was to use several models from several vendors not just Claude. Also the interesting part is the developing pipeline I have had setup as cicd with my own tool impactguard whitch saved me a couple of times and hard rules on the agent.md(70% to 80% of those rules i wrote them at hand). Yes the code and the readme might look like ai slop as pointed out trailer but is efective at finding bugs. Also we found earlier another bug with the fuzzer https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/23945. I welcome constructive criticism and feedback. Any input is useful to me