Skip to main content
Control and respond

Alerts and the inbox

AI Governance raises an alert when it notices something a person should hear about. This page says what raises an alert, what the inbox shows, how to work an alert and which alerts stay in the Portal.…

Written By Christopher Scaminaci

Last updated 4 days ago

AI Governance raises an alert when it notices something a person should hear about. This page says what raises an alert, what the inbox shows, how to work an alert and which alerts stay in the Portal. To send alerts elsewhere, see Alert channels, policies, quiet hours and digests.

At a glance

  • What it does: tells you about findings, new tools, quiet sources, expiring tokens, name requests and more.
  • Plan: every plan has the inbox and email. Other channels need Pro or higher.
  • Who: Administrator and Reviewer read the inbox. A Reveal approver sees only the alerts about name requests.
  • Where: AI Governance → Alerts → Inbox. The menu shows a badge for open alerts.
  • Needs: nothing. Alerts raise on their own.

What raises an alert

An alert is raised when StackJack notices something such as:

  • sensitive content in an AI prompt or answer,
  • a call that was blocked or warned,
  • a source that went quiet or lost coverage,
  • an ingest token about to expire or expired,
  • a request to reveal a person's name,
  • one of your own storages needing attention,
  • a client setting that differs from its standard, or a purchase or seat cleanup that stopped,
  • one of your own alert channels failing.

The kinds, by the category names the inbox filter uses:

CategoryAlert kinds
Sensitive contentSensitive data found, Sensitive data blocked, Sensitive data warned about, Possible sensitive data to review (AI not sure)
Source healthSource silent, Collection gap, Pull failing, Scan coverage lost, New attribute names, Ingest token expiring, Ingest token expired, AI spending budget threshold reached, AI request refused by StackJack, MCP proxy server failing, Claude hooks secret rejected, Browser extension cannot read an AI site, Browser extension token refused, Endpoint agent out of date, Endpoint agent coverage gap, Microsoft partner app needs attention
PrivacyEvents refused before the staff notice, Name reveal requested
InventoryUnapproved AI tool, Denied AI tool in use, Unmatched identity, Approval requested, Vendor change failed, Ticket requested for an AI finding, Agent rules acted on AI calls, Change approved without a person, Setting differs from the standard, Enforcement stopped for a setting, Automatic purchases stopped, A purchase needs checking, A purchase price changed, A purchase went through after its run stopped, Seat cleanup stopped for a group
IncidentsIncident opened, Incident clock due
ArchiveArchive target needs attention, Archive backlog, Archive data loss, Archive conflict, Archive credential expiring, Storage buffer full, text capture paused
ChannelsAlert channel failing

The inbox

Every alert is in the inbox, whether or not a channel also sent it. Channels are extra. They put the alert where your team already looks.

The list shows the Severity, the Alert, When and Status. Filter by Status, Category and Severity.

An alert has a status:

  • Open. New.
  • Acknowledged. You have seen it. Select Acknowledge.
  • Resolved. It is dealt with. Select Resolve.

Select Reopen to open it again. Times show in your organization's time zone, which you set on the Schedule tab of Alerts → Alert settings.

Open an alert

Select an alert to open it. Every kind of alert has the same page.

The top says what happened in one sentence, for example "TP-LT-0142 stopped reporting on Nov 13. AI use on it is not recorded until its agent reports again." One button is filled in: the next thing to do for that kind of alert, such as Open the computer, Approve tool or Open the activity.

The main column shows the facts first and the advice last:

  • Who and where, and what triggered it. What you see depends on the alert:
    • A coverage gap or an old agent version shows the computer: its name, its person, what changed in words, its last report and how long ago, the agent version and the operating system. It also says how many other computers have the same problem now. An alert that stands for many computers lists them, and counts the list as it is now.
    • An unapproved AI tool shows the tool, the first time StackJack saw it with a link to that activity, what saw it, the source, and the people who used it with the computers their use was seen on.
    • A rule match shows the rule's name, its pack and its mode, where in the text it matched, the person, the computer, the app, and the client named in the text. The matched text itself is never shown here.
    • Every other alert shows the thing it is about: the source and what went wrong, the key, the channel, the incident and its deadline, the storage, the budget, the change.
  • Findings behind this alert lists where a rule matched, with the masked value.
  • What happened lists the events in order: when it happened, when the alert was raised, when it was sent, and who or what resolved it.
  • What StackJack did shows each action a rule asked for and whether it was done.
  • What to do says what this kind of alert means and lists one to four steps, each with a link to the page where it is done.

The side column shows the bare facts (the kind, the category, your organization, the times and whether it was seen before) and Who was told: each policy and channel, the result and why. When no policy sends this kind of alert anywhere, it says so in one line.

A computer's name is shown only where you may see real names in that organization. Otherwise you see "Computer" and a short id. Where StackJack did not record a fact for an older alert, the page says "Not recorded for this alert".

StackJack closes a coverage gap alert by itself when the computer reports again, and an old agent version alert when the computer updates. The page then says StackJack resolved it.

The other buttons at the top depend on the alert:

  • Approve tool and Deny tool appear on an alert about an AI tool, for an administrator. Either one sets the tool's approval, the same as on the tool's own page, and resolves the alert. If the answer needs more (for example, denying the last approved MCP server), the page says so and the alert stays open.
  • Acknowledge, Resolve and Reopen change the status.
  • More holds Open incident from this alert. It opens an incident linked to the alert, and to the activity, finding or tool it names, and takes you to it. It is for an administrator or a reviewer, on the Business plan and above.

Work many alerts at once

Tick the alerts you want in the inbox (or Select all on this page), then select Acknowledge selected or Resolve selected. Each change is recorded on its own, as when you change one alert. The inbox says how many it changed and how many it could not.

On a row about an unapproved tool, the line under the title names the tool and how many people used it, for example "ChatGPT (OpenAI), 3 people".

The alert that stays in the Portal

One kind of alert stays in the Portal inbox and goes nowhere else: "Possible sensitive data to review (AI not sure)". The AI check was not sure about the content and asks a person to look. It is never sent by email, Teams, Slack or a ticket. No policy can route it. A doubtful answer does not wake your team. Its severity shows at most Medium.

Who sees which alert

RoleAlerts
AdministratorEverything: the inbox, channels, policies, schedule.
ReviewerThe inbox, and the settings read-only.
Reveal approverAlerts about name reveal requests only.
AuditorThe settings, read-only. No inbox.

An alert shows a person by pseudonym unless your name policy and the role allow names.

What an alert contains

An alert sent outside StackJack contains its kind and category, its severity, the rule's name and pack, the part of the exchange it was found in, the actions requested and taken, how many matches, the source's name and kind, your organization's name, the person's label, the client's name, counts and dates, and a link to the Portal.

An alert never contains the matched value, any prompt or answer text, a masked token, a tool argument, the reason or case reference of a reveal request, a vendor's message or an error text. Those stay in StackJack.

Troubleshooting

What you seeWhat to check
"No alerts match."Clear the filters.
"You no longer have a role that can see these alerts."Your role changed. Ask an Administrator.
"You do not have a role that can change this alert."Only the roles in the table above can change an alert.
Alerts reach the inbox but not Teams or emailCheck your policies. An alert that matches no policy stays in the inbox only. See Alert channels, policies, quiet hours and digests.
Too many alertsMake noisy rules Test mode, or use a digest policy.

Next