[Article] Exploiting Thoughtcrime in LLMs

New malware variants seem to include source code comments about how to build WMDs, presumably in an effort to hamper analysis by AI-based tools. I find myself wondering if the same technique can be used by authors to "protect" their work, too...