Human Review of Private AI Conversations

After earlier reporting about human review of ChatGPT and Anthropic conversations, 404 Media now reports contractors reviewing Copilot prompts and uploaded images. A different product makes the exposure boundary recur, but the new item still comes from the same reporting organization and reaches the corpus through Bluesky. This strengthens the watch line without proving industry prevalence or meeting the independent-source publication threshold. Check provider controls, retention policies and an independent investigation next.

A new report says humans review ChatGPT conversations used for model improvement, including potentially sensitive prompts. This is more than a privacy complaint: it points to a recurring operational boundary between model improvement and confidential use. Keep watching for disclosure controls, customer-owned retention, and enterprise products that separate evaluation data from private conversations.

Reporting on Project Lily says human reviewers read selected ChatGPT conversations, with the same reporting pointing to a similar practice at Anthropic. This may create a distinct enterprise market for verifiable no-human-review processing, review disclosure and routing of sensitive conversations. It remains a weak signal because most publications trace back to one investigation; independent policy changes or enterprise procurement requirements would confirm it.