More or less trained LLMs are obfuscating "stolen" training data. The status quo used to call this sort of thing theft and piracy