What people sent to AI tools, inspected before the model saw it
Worth a look
Loading activity...
Time
Person
Tool
Surface
Detected
Outcome
Rule
No activity matches these filters.
Responses
What models sent back, inspected before the person saw it
An allowed prompt does not guarantee a safe answer. Every response is inspected — including
responses to prompts that passed cleanly.
Time
Person
Model
Prompt outcome
Found in response
Outcome
Reason
No responses match these filters.
Stats
Shape of activity across the selected period
Volume over time
InspectedBlocked
What is being detected
Where it is happening
Highest block rates
click to open a profile
A high rate usually means the rules do not fit that person's role, not that they are doing
something wrong.
What gets flagged
by category
Tools used
by outcome
Recent activity
Time
Tool
Surface
Detected
Outcome
Rule
Framework coverage
Controls evidenced by enforcement activity
ISO/IEC 42001 mappings are provisional.
Control references have not been checked against the published standard. Treat NIST AI RMF
figures as reportable and ISO figures as indicative until validation completes.
NIST AI RMF
ISO/IEC 42001
Policy rules
Evaluated in priority order. The first match decides the outcome.
Priority
Rule
Applies to
Who
Tool
Category
Outcome
Status
No rules are configured. Content is allowed by default, and file uploads are blocked until a rule allows them.
Devices
Managed computers running the device agent, and whether each one is working
Healthy: reporting, with its certificate and proxy setting in place.
Silent: no report recently (stopped, offline or blocked).
Stopped: the agent said it was stopping.
Tampered with: its certificate or proxy setting was removed or changed.
Loading devices...
Loading events...
This device has reported no events.
Person
Computer
System
Health
Last report
Checks
Recent events
No devices match these filters.
Browser extensions
Installs of the browser extension that checks prompts typed into AI sites
Silent: no report recently. The extension may have been disabled or removed, or the browser
is closed. An install that was never made does not appear here. No page, address or prompt is ever
reported, only the browser, version and when it last checked in.
Person
Browser
Version
If the gateway is down
Health
Last report
No browser extensions have reported yet.
AI app use
Which AI services managed devices connected to, from how many devices, and from where
These are counts of connections, not of prompts. The device agent notes that a device connected
to an AI service and which kind of program opened the connection. It cannot see what was sent.
Browser connections are checked by the browser extension; other app connections (a desktop
app) cannot be inspected, and to stop one you block it with your device management tool.
Unidentified means the program could not be told, or the agent is too old to say.
Names are hidden. Devices are shown by hostname, and people are counted but not named. Which AI
services a named person used is personal data; an administrator can switch names on
(AI_USAGE_SHOW_PEOPLE) once that is approved.
Loading usage...
AI service
Devices
People
Connections
Per device
Browser
Other app
Unidentified
Last seen
Person
Device
AI services
Connections
Browser
Other app
Unidentified
Last seen
No connections to AI services from managed devices in this period.
Blocked wording
What a person sees when a model's answer is blocked
The wording never reveals what was blocked. It can only name the kind of content, using
{category}. The fallback applies to any category without its own wording.
Loading wording...
Category
What the person sees
Source
Preview, as the gateway renders it
Team limits
Which teams each analyst may see
A limit narrows what an analyst sees to the teams you name. Compliance and admin roles always
see everyone, so they cannot be limited. An analyst with no limit sees everyone; an analyst
limited to no teams sees no one.
Loading people...
Person
Role
Their team
Can see
Legal holds
Freeze one person's records against the retention purge
While a hold is active the daily purge skips everything this person owns, whatever its age:
their inspections and response checks, devices and their events, browser extension installs
and console access records. A hold covers one person, not a period or a team. Releasing it
lets the normal retention rules apply from the next run.
Place a hold
Loading legal holds...
Person
Case reference
Placed by
Status
Model trust
Reliability, drift and bias for each model endpoint. The signals are never blended into one score.
Browser-extension and device-agent traffic is not covered by trust scoring. A signal that
has not been measured is shown as such, never as OK.
Loading trust signals...
Endpoint
Overall
Coverage
Reliability
Drift
Bias
Open alerts
Console access
Who viewed governance data, and what they looked at
Your own activity in this console appears here too. Anyone with access to this page can see
which records you opened.