← All discussions

Other Discussion

Barclays AI credit scoring rollout hits explainability roadblock

Mia Roberts · 19 Aug 2026

Our compliance crew just received the FCA's July 2026 guidance on algorithmic transparency, and the AI‑driven credit scoring tool we launched two weeks ago is suddenly under fire. The model's black‑box nature means we can't easily pull out the factors behind each rating, and the regulator's new audit requirements are leaving us scrambling for documentation. Our data science team says they'd need weeks to build a transparent layer, but the business side is pushing to keep the tool live because it’s already boosting approval rates. Has anyone managed to retrofit explainability into a live AI credit system without pulling the plug, and what practical steps did you take to satisfy the FCA?
👍 539 💬 4

4 replies

Hugo Williams · 19 Aug 2026

I tried plugging in a SHAP dashboard for the risk team, but they said it still wasn't granular enough for the FCA.

Fernando Richardson · 19 Aug 2026

Our biggest blind spot is the audit trail for model versioning. If regulators demand proof of every tweak after a data refresh, we’ll have nothing to show and could be forced to shut the tool down, so we should get a version‑control system like MLflow or DVC in place ASAP.

Pedro Higgins · 19 Aug 2026

Honestly, the model could be silently drifting as market conditions shift. If we don’t set up real‑time monitoring, the scores we ship tomorrow might be out of sync with the FCA’s fairness expectations and could land us in a bigger compliance mess.
C

Carolyn Marshall · 19 Aug 2026

What we haven’t flagged yet is that the training data includes legacy credit bureau fields that are now considered sensitive, so any explainability output could inadvertently expose protected class information and trigger discrimination claims. We've got to clean that up now or we could end up with hefty fines on top of the audit headache.

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