Overview and Data Insights
The Overview and Data Insights pages summarize what Sanitized AI caught and how people responded. Admins see the whole organization. Managers see the people who report to them. For who sees what, see Roles and access levels.
Both pages share one time range, which you pick at the top of the page: the last 24 hours, 7 days, or 30 days, or all time.
Overview
The Overview page is your daily check. The top row shows:
- Events: how many prompts and files were flagged in the period.
- Overridden flags: how many flagged prompts people sent anyway.
- Active guardrails: how many of your guardrails are turned on.
- Guarded sites: how many AI tools are protected.
- Users, or Your team for managers: how many people are included.
Below that, activity cards show events over time, the share of prompts that were flagged (Flagged vs clean), the most recent events, and the guardrails that fire most. If your organization records AI usage, an AI Usage card shows how many prompts people sent and on which tools. Its Active Users figure counts people whose extension checked in during the last 14 days, whatever time range you pick, the same rule as the Usage column on the Users page.
What people use AI for
The AI Usage page can also show what your organization uses AI for, by task, such as Summarizing or Coding, and by topic, such as Legal or Finance. It also shows which tasks or topics are flagged more often than their share of prompts.
It’s off until an admin turns it on: in My Organization > Data & privacy, turn on Count prompts by task and topic in the What people use AI for card. Only daily counts for the whole organization are kept. Nothing is kept about who sent a prompt, and the prompt itself is never stored. The labels are estimates and never change whether a prompt is warned or blocked. To learn how they’re worked out, click How is this worked out?
Data Insights
Data Insights goes deeper. Its cards are grouped into five sections.
At a glance
- Risk Score: a score for your organization or team, compared with the previous period. To learn how it’s calculated, click the information icon next to it.
- Flagged events for the period.
- Threat Velocity: whether flagged events are rising or falling.
- Active Users: how many people are using Sanitized AI, and how many of them triggered a flag.
People and teams
- Risky Users Leaderboard: people ranked by risk points. Turn on Repeat overriders only to find people who often send flagged prompts anyway.
- Department Risk Profile: risk by department. Departments come from each person’s profile on the Users page.
- User Risk Distribution: how many active people fall into high, medium, and low risk.
What was exposed
- Risk Activity Breakdown: responses and top guardrails for each category of data.
- High & Critical: the mix of severities flagged.
- File Uploads Flagged: files that a guardrail caught.
- Largest exposures: how many different sensitive values each flagged prompt or file held, and which AI tools and departments the ones with 50 or more came from. A large one usually means a list was pasted or uploaded. The card appears once events have a recorded size.
Which AI tools
- Top Risky AIs: the AI tools where most risk happens.
- Coverage Gaps and New This Period (admins only): which tools and guardrails are new or not yet covered.
How people responded
- Risk Prevented, High-Risk Actions, and Flags Ignored: how many flags ended in a safer prompt, were sent anyway, or were dropped.
- Response Trend: how people responded to flags over time: sanitized, re-edited, overridden, or ignored. Click a response to show only that one.
- Enforcement Effectiveness (admins only): how people respond to warnings and blocks.
To download a table, click CSV above it.
How scores work
Every flagged event gets a severity from 1 to 5, based on the kind of data found:
| Level | Covers |
|---|---|
| 5, Critical | Secrets that grant direct access to systems, accounts, or funds |
| 4, High | Government identity documents that could be used for impersonation |
| 3, Medium-High | Regulated personal, health, and financial records |
| 2, Medium | Confidential business and internal information |
| 1, Low | Everyday contact and network details |
The scores on Data Insights build on that severity:
- Risk Score runs from 0 to 100 and reflects the average severity of the flagged events in the period: 0 to 25 is low risk, 26 to 70 is moderate, and 71 to 100 is high.
- Threat Velocity compares the number of events with the previous period of the same length. A positive percentage means events are rising.
- Risk points add up for each person. Each flagged event adds its severity, plus one extra point when the person sent it anyway, up to 5 points per event. More than 50 points is high risk. 10 points or more, or more than 2 events, is medium risk.
For the full explanation in the dashboard, click the information icon next to any score.
Drill into a department or a person
- A department: in Department Risk Profile, click a department. Its page shows flagged members, flagged prompts, the override rate, recent activity, and the AI tools involved.
- A person: click a name in Risky Users Leaderboard, on a department page, or on My Team. Their page shows their flagged events, override rate, risk points, rank in their team, behaviour over time, any exceptions that apply to them, and flags they disputed. To download their events, click Export events.
If you use the extension yourself, My Activity shows the same page for your own activity. Nobody else’s activity appears there.
Download a report
- On Data Insights, choose the time range.
- Click Report.
A PDF named AI Usage Risk Report downloads. It includes the key figures, the severity mix, coverage, and the top people and guardrails for the period. It’s marked Confidential: share it with care.
For a report mapped to the laws and standards that apply to you, admins can click AI Safeguards Report, which opens the Compliance tab of My Organization. The button appears once Sanitized AI turns on the report for your organization. See Compliance and the Audit Assistant.
Monitoring-only events
Events that were recorded without showing a warning are kept separate. To see only those, turn on Monitor only at the top of Data Insights or Events.