← All discussions

Technology & Data

Procurement signed AI contracts with 30-day logs, legal wants seven years

Grant Shaw · 3d ago

Procurement leads at a UK insurer just found our HR screening vendor keeps model logs for 30 days, while legal wants seven years of evidence for the August high-risk AI deadline. The renewal clause puts audit rights behind a £250k tier, and we can't get the vendor to confirm which model version screened each candidate. We're now rebuilding an evidence trail from ATS exports, and it's not pretty. Are other tech and data teams rewriting AI contracts now, or waiting for the regulator to force the issue?
👍 594 💬 4

4 replies

J

Jerry Mitchell · 3d ago

Honestly, trying to get a seven year evidence dump from a vendor that only keeps 30 days is a wild goose chase. I’d push for an addon that writes immutable logs to Snowflake and then archives them to S3 for at least 180 days.
L

Lily Hart · 3d ago

What we haven’t flagged yet is that the 30‑day window could erase any audit trail before the regulator even asks for it, leaving us exposed to hefty fines for missing evidence. If we can get the vendor to push raw logs to an immutable bucket like AWS S3 with versioning, we’ll at least have something to show even if their internal store expires.

Quinn Mendoza · 3d ago

My mistake was signing that renewal without running a test SAR first, and now we've got no version data to fall back on.

Iris Russell · 3d ago

We've run into the same snag and ended up adding a separate data retention addendum that forces the vendor to keep immutable logs for at least three years. The trick was to tie the addendum to the existing service fee so we didn't have to hit the £250k audit tier, and we now pull the version info straight from the model registry we set up in MLflow. If you haven't tried a lightweight registry yet, it can give you a cheap audit trail and saves you from stitching together ATS exports later.

Log in or create a free account to join the conversation.