Claude Found Hundreds of Firefox Bugs That Decades of Fuzzing Missed

Large language models can automatically generate crashing inputs and surface logic errors across large codebases, finding many bugs that decades of fuzzing and static analysis missed. In short tests, an LLM produced hundreds of unique crashing inputs and identified distinct classes of logic bugs beyond conventional fuzzers' reach. — If LLMs routinely uncover longstanding, high‑severity bugs in widely used software, that changes how vendors, open‑source projects, regulators, and attackers approach software security, liability, and disclosure practices.