Security reporting3 min readField note 17

How to count sensitive AI pastes when one paste matches several rules

Work through a Claude paste and a Gemini paste on one Windows device. Reconcile five detector matches with two gestures before reporting incident counts.

Northstar Sensitive pastes workspace with a sample receipt and its detector categories.
The idea in focus

A reviewer can count the observed gestures and classification matches independently in the fictional Northstar workspace.

Field note / 17
In brief

Two sample pastes produced five matches. Each number answers a different question.

  • Use distinct event IDs to count observed gestures within a tenant.
  • Add the findings for each category to count pattern occurrences.
  • Use the review process to decide how many incidents to report.
In this article

Choose the unit before opening the report#

A security manager asks how often sensitive text was pasted into an AI tool. Define your reporting window, tenant, enrolled device population and supported browser scope. Count distinct paste receipts with a finding for the observed-gesture total. Keep a second column for pattern matches and a third for incidents your team has reviewed. The two receipts in this exercise are a selected example from NS-WIN-014.

Northstar Sensitive pastes search lists receipts matching the API-key category.
A category search finds matching receipts. A sample built from that search excludes events that only matched other categories.

Read the three matches on Claude#

Open Sensitive pastes and search for Claude. The local detector found one key-shaped value and two email occurrences on EVT-1051. Counting the findings as separate employee actions would turn one observed paste into three. Category totals alone cannot tell whether the key was valid, whether an address repeated or whether the employee sent the prompt.

One Claude paste receipt lists one API-key pattern and two email-address matches.
EVT-1051 contributes one gesture and three matches. Whether the two email matches refer to different addresses is unknown.

Read the second receipt on Gemini#

Search for Google Gemini and open the sample from the same hostname. It carries a different receipt ID and its own observation time. Keep it as a separate event even though the device and one category match the Claude example. A filename, hostname or detector label makes a poor deduplication key.

The Gemini receipt lists one email address and one labeled-secret finding.
EVT-1054 adds one gesture, one email occurrence and one labeled-secret occurrence.

Reconcile the totals in a small table#

Within these two receipts the distinct category labels are Email addresses, API key and Labeled secret. The same email label appears on both rows. Report the measure alongside its number. Adding category counts gives five matches. Counting distinct receipt IDs gives two observed pastes. Counting affected devices gives one.

Worked table totals two paste receipts and five matches across three category labels.
The two selected rows contain three email matches, one API-key match and one labeled-secret match.

Handle overlapping exports and genuine repeats#

Suppose a second export overlaps the first window by an hour. Combine records by tenant and stable event ID, retain the export dates and check any conflicting fields before counting. Two separately observed gestures can carry identical category counts. Preserve both when they have distinct receipts, then let an analyst decide whether they belong to one case.

An overlapping export repeats EVT-1051 while the merged sample still contains two unique receipts.
The repeated export is a worksheet example. The stable tenant and receipt identifier let a reviewer reproduce the total.

What can management take away?#

For the worked example, report two observed pastes with findings on one managed device and five detector matches. The case outcome is pending review. For a weekly report, include device coverage, inspection gaps, scope changes and the count of reviewed cases alongside the event numbers. Raw paste totals cannot establish individual productivity, confirmed data loss or the number of unique secrets.

Reporting worksheet separates observed gestures, classification matches and unresolved case outcome.
This summary describes two selected receipts. A fleet rate requires a defined reporting period and measured device coverage.
Read the primary sources

Sources & further reading

  1. Inspect the two sample receipts
  2. Sensitive-paste investigation workflow
  3. How to read activity evidence
← Back to all articlesBack to top ↑
A CLEAR PLACE TO START

Plan your AI visibility pilot.

Create a brief for your team. Your answers stay in this browser and are included in your download.

AI tools in scope
More supported tools

This creates a planning document. It does not enroll devices or submit a request.