The weights contain malware? What security vulnerability do they contain? Is the claim that they will write insecure code? This seems easy enough to test - run a sufficiently large sample through another LLM