palantir and nvidia have stopped using claude/GPT when dealing with proprietary information due to concerns their data is being used to train a model that replaces them.
instead they're using claude + gpt to automate generic HR, finance, legal work then shifting to open source + air-gapped private servers for confidential work (codebases, drug designs, software).
anthropic, openai's argument is they need the data to prevent misaligned use.