A pattern for AI audit trails that keeps privacy and proof apart: don't make the guardrail layer store prompts and completions. Per request, record validator verdicts in order, rule IDs, a hash of the active policy config, SHA-256 digests of request and response, and an HMAC signature. If you keep payloads elsewhere under your own retention rules, the digest proves they match. #LLM #AIGovernance #Privacy #MLOps
Related
Writing is thinking. It's one of the reasons I started blogging again and has helped me immensely. Can't recommend it st...
Writing is thinking. It's one of the reasons I started blogging again and has helped me immensely. Can't recommend it strongly enough.AI is still ok for proofreading and mundane ta...
Dew Drop - November 10, 2026 (#4789)https://www.alvinashcraft.com/2026/11/10/dew-drop-november-10-2026-4789/#dotnet #web...
Dew Drop - November 10, 2026 (#4789)https://www.alvinashcraft.com/2026/11/10/dew-drop-november-10-2026-4789/#dotnet #webdev #ai #csharp #windowsdev #cloud #appdev #visualstudio #gi...
"In defense of the human code review"I briefly posted a version of this earlier, but I've retitled, reframed and updated...
"In defense of the human code review"I briefly posted a version of this earlier, but I've retitled, reframed and updated it.It's about the increasing pressures to speed up or repla...