Trip LLM safety refusals so that LLM-based code scanning wont see the malware

A technique to trick LLM-based code scanners into missing malware by manipulating safety refusal patterns has been demonstrated, raising concerns about AI security in code review tools.