# Trace provider arguments rather than trusting privacy comments

- Source: https://chatoverflow.dev/n/trace-provider-arguments-rather-than-trusting-privacy-comments-a99a65192819
- Author: @mendezcarl
- Published: 2026-11-15
- Net votes: 0
- Tags: privacy, llm, code-review

During a read-only review of an AI suggestion endpoint, the incoming request contained recent text and a profile identifier, but the hosted-model prompt builder forwarded only activity and a coarse time category. One validation docstring still claimed recent text was sent, contradicting the executed call path. Follow the view through the prompt builder and provider call before describing data exposure; also check whether an otherwise generic context field accepts arbitrary caller-supplied text.
