Evaluate Shadow AI
See which AI services are used in your organization, classify them, and export the result as evidence.
The Browser Guard reports visits to AI services and the checks of prompts in the AI chats. In Shadow AI you see this data and decide which AI services may be used in your organization.
Where you find it
Open Shadow AI in the sidebar. At the top right you choose the Period: 7 days, 30 days or 90 days. It applies to the tabs Services and Prompts in the browser and to every tile that names it. If your plan includes the export, Export report opens the Reports tab.
What you need to do
When there is something to decide, the box What you need to do stands below the head of the page; otherwise it is left out.
- Services are not yet classified: the box names up to three of them. Classify now opens the Services tab, filtered on Not yet classified. As long as a service is not classified, users get no notice there that it is not released.
- Services are suspected to be AI: Review opens the Suspected AI tab.
- If the classifications could not be loaded, the box shows no number and says so; Retry loads again.
The tiles
| Tile | What it counts | A click opens |
|---|---|---|
| Visited domains | The domains of AI services visited in the period, the same rows as the table on the Services tab | the Services tab |
| Classified as not released | The domains you classified as Not released, in the whole project, regardless of the period | the Services tab, filtered on Not released |
| Checked prompts | Prompts whose content the Browser Guard checked in the period, only in services with content checking | the Prompts in the browser tab |
| Blocked or made unrecognisable | Checked prompts that were blocked or made unrecognisable, with their share of all checked prompts | the Trace Explorer |
If the answer carries no number, the tile shows Not known, never 0.
The tabs
| Tab | What you do there |
|---|---|
| Services | See and classify the visited AI services |
| Suspected AI | Confirm or dismiss findings of Palveron Discover; only if your plan includes Palveron Discover |
| Prompts in the browser | Look at the checks of prompts from the browser |
| List of all known services | See the list the Browser Guard compares visits with, and add your own domains |
| Reports | Create the Shadow AI report (CSV); only if your plan includes the export |
Classify AI services
The Services tab shows a table with the columns Service, Kind, Classification, Visits, Checked prompts and First seen. Services not yet classified and new services are at the top.
Above the table you filter on one line: the search field Search service or domain … and the select fields Any classification and Any kind. A field that is set shows its kind with the value, for example "Classification: Not released". When a filter is set, Reset filters appears.
In the Classification column you choose for each service:
- Released: the service may be used.
- Not yet classified: there is no decision yet.
- Not released: the service may not be used.
Only an Admin can classify; for other roles the choice is locked and names the reason. For a service that is not released, the Browser Guard shows a warning; the person can continue, and continuing is reported. If the column shows Status not evaluable, a status this page does not know is stored for the domain; classify it again.
What the numbers mean:
- Visits count in the selected period, only on devices with Browser Guard.
- Checked prompts counts per service across all its domains and appears on the domains where the Browser Guard checks the content. A dash means it checks only the visit on that domain, or the number is not known; the hint on the cell says which.
- First seen names the first visit when it was within the last 7 days; otherwise it says "more than 7 days ago".
If more domains were visited in the period than the table shows, a line below it says how many of the most visited domains are shown.
Suspected AI
The Suspected AI tab shows the findings of Palveron Discover: services outside the list of known services that behave like AI. While Palveron Discover is switched off, the tab offers Activate; when it is on, Deactivate switches it off again. Editors, Admins and Owners can switch it on and off.
Above the table are the figures Pending Review, Confirmed AI and Dismissed. The Vector column names the kind of finding:
- Response pattern: the answers of the service resemble those of known AI services.
- Extension: an installed browser extension with AI was detected. Show details shows, among others, Trains on user data. The device reports these details; they are not verified.
For each finding, decide with Confirm AI or Dismiss; only an Admin can do this. Confirming adds the service to your list as not released; the Browser Guard warns once it has loaded the new classification. Hide dismissed keeps the list short.
Prompts in the browser
The Prompts in the browser tab shows the prompts whose content the Browser Guard checked in services with content checking, in the selected period. At the top are Platform Distribution, Decisions over time and Personal data and credentials intercepted.
The table Recent Browser Activity shows each check with Time, Platform, Prompt, Decision, Detected data types and Transmission. "Sent" in Transmission means handed over to the service, as reported by the browser. View all in Trace Explorer opens the full list.
If it says No prompt-level data yet, the Browser Guard is not installed or has not been used in an AI chat yet; see Install the Browser Guard.
List of all known services
Under Your custom domains, Admins use Add domain to add internal AI services or industry-specific platforms, with Domain, Display name and Category. Entries apply to this project only. The Browser Guard refreshes its list on its own; it detects new domains within 24 hours.
Below is the Directory of AI services, the list the Browser Guard compares visits with. You filter on one line: the search field Search AI services or domains..., the select All categories, the select All origins (only when an entry carries an origin) and the level All levels, Checked or Automatically detected. Automatically detected domains take effect in detection but are not checked yet. The list is shown page by page, with Previous and Next.
Export as evidence
On the Reports tab you first choose the Report period: All time, Last 7 days, Last 30 days, Last 90 days, Current quarter, This year or Custom with From and To. It is independent of the period at the top of the page.
Generate CSV creates the Shadow AI report (CSV): the checked prompts, the detected visits to AI services and the reports of the Browser Guard, merged chronologically and marked with each domain's sanctioned status. The reports include warnings and the choices made, blocked sends, sends not recognized as a prompt, sends that could not be checked and answers before an upload; the Source column names the kind of each row, including when a person continued despite a warning or despite the notice on outdated rules.
Five columns at the end of each row concern the reports of the Browser Guard:
- Number of occurrences: one row can stand for several equal occurrences. For evidence, count this number, not the rows. Empty on rows that carry no number, such as checked prompts and visits.
- Service ID: the identifier of the AI service the Browser Guard reported; the Tool column shows its name.
- Reason not checked: for a send that could not be checked, why: "No answer from the gateway in time", "Gateway not reachable" or "Gateway did not process the check".
- Device time (stated by the device, not verified): the time the device states for the event. It is a claim of the device; the period is decided by the Timestamp, the arrival at the gateway.
- Device time compared with arrival: "Consistent with arrival", "More than 5 minutes after arrival" or "More than 7 days before arrival".
A field without a value stays empty; a value the dashboard does not know appears as its identifier. An export holds at most 10,000 rows; narrow the period for complete data.
A review routine
- Once a week, open Shadow AI and read the box What you need to do.
- With Classify now, classify every service still in Not yet classified.
- If Palveron Discover is on, decide every pending finding under Suspected AI.
- Under Prompts in the browser, look at prompts that were blocked or sent after a warning.
- Once a month, export the CSV report for your compliance documentation.
What the data does not show
- Only visits to AI services on the list are reported, plus redirects of an AI chat to another domain and the findings of Palveron Discover. Other browsing is not reported.
- The Browser Guard does not tell whether a person uses a service with a private or a company account.
- Prompts are checked only in the AI chats; on other AI services only the visit is reported. See the overview.