LLM generated content is not the user's, it's the company's, licensed to the user under ToS, not copyright. Even more so with the raw proprietary reasoning traces, which most users never see and are only available as summaries created via a secondary LLM, or sometimes not at all. So the proprietary reasoning trace can never belong to the user due to ToS and the user never seeing it.

Many companies don't believe ZDR is real, so they log LLM conversations and tell employees to never share sensitive data with an LLM in the first place. This openai incident might justify it.