Signed in asM. HalloranCompliance officer · Compliance/BSAThe console in full: every actor, every event, the adjudication queue, and the actions on the pending policy proposal.
Monitor pilot · week 10AI Data Guard — Alder Creek Community Bank$850M · Ohio; 11 branches · 243 staffTrailing 30 daysJul 16 – Aug 14, 2026
What AI use carried, and what happened to it

1,144 detections in 3,805 AI interactions — 302 stopped, 337 transformed, 505 sent as typed

Trailing Jul 16 – Aug 14, 2026. 30.1% of AI use carried a sensitive class — discovery counts every interaction, detection counts the subset, and the two are never divided into one another without saying so. Every figure below opens the events behind it, filtered to exactly what it counted and over this same window.
What happened to those detections

A detection is not an exfiltration. Blocked, redirected and abandoned submissions reached no destination; transformed ones reached it with every detected span masked or tokenized and the original text withheld; overridden and monitored ones went as typed.

An event can carry several classes, so class instances outnumber detections: 1144 detections in this window carried 1847 class instances, of which 1840 fall in the six classes the tiles draw. A loan application carrying a name, a tax identifier and an account number is one detection and three instances. The two figures are never divided into one another.

Top AI tools detected

Where AI happens here

Ranked by interactions over the window. Severity is an assessment somebody performed, recorded on the registry — which is why an unassessed tool carries no chip rather than a green one. Each row opens the detections behind it — the events that carried a data class, which is the middle figure on the row; interactions are the discovery denominator and carry no chain of their own.
The pilot, week by week

Usage and detection, kept apart

Each dated rule is a policy version taking effect. Historical events carry the version that decided them, so tightening a policy never rewrites what came before it.
03356691,004v2v3v4v512345678910
AI interactions — every use, sensitive or notDetections — the subset carrying a data classPolicy version in force from that week

Weeks 1 to 10 of the pilot, Week 1 to Week 10.

Findings

What the log says, in four quadrants

Adverse and favorable looking back, risk and opportunity looking forward. Every one of these is computed from the event log and drills to the events behind it — a control product that only ever reports bad news teaches its buyer to stop opening it.
Week 10 brief

Week 10 compliance brief

Week 10: Loan Files up 31%, override rate 11%

Loan Files rose 31% week on week, to 291 events in the window. The override rate is 11% in Week 10 against a 15% target, 11 points below its Week 5 rate.

Written by ordinary code from the same figures — the designed fallback, not a degraded mode Across the trailing 30 days the log records 1,144 detections in 3,805 AI interactions, 30.1% of AI use; 302 of those detections reached no destination and 505 went as typed.
The sanctioned assistant

Ask

The internal assistant the coach card redirects staff to, governed by the discipline this product sells: it cannot produce a figure, only ask a tool for one. Answered from the figures above, or not at all — and answered inside the scope of whoever is signed in, because a chat that reached past the permission model would be the way around it.