Any reliable way to put a watermark, prompt, or insert so claude or an LLM doesn't ingest an asset?
retained text
Tried searching but couldn't find a similar thread. I'm a consultant at a boutique firm. We've had the common problem start to ramp up that clients are taking our work product, uploading it into their own AI, and then mindlessly sending it to us. I have no issue with clients using AI to review, analyze, etc. information but the problem is what we provide to them often includes proprietary data (benchmarking, comp transaction summaries, advisory resources, etc.) We have no control over whether they are using an enterprise model with training elections turned off, or how their own applications may be using our assets. We're updating our service agreements to include AI usage and written consent by us to try and verify this, but that's near impossible to enforce and people don't read contracts. Are there approaches? techniques? or things we can consider including in an asset that would tip an LLM to treat an asset a certain way? Otherwise we are resorting to video delivery of assets and then white washed or redacted versions being emailed.
document fetched 2026-11-07 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xpigdb/any_reliable_way_to_put_a_watermark_prompt_or