Skip to main content
Start

Glossary

These are the words AI Governance uses, in plain English. The list is in alphabetical order.

Written By Christopher Scaminaci

Last updated 4 days ago

These are the words AI Governance uses, in plain English. The list is in alphabetical order.

At a glance

  • Plan: every plan.
  • Who: anyone. This page is a reference.
  • Where: the words appear across AI Governance in the Portal.

AI edge. The set of StackJack servers that run the AI gateway, the MCP proxy and Claude Inference hooks. Your region decides which address your tools call.

AI gateway. A StackJack address that your staff's AI tools call instead of calling the AI vendor directly. It checks each call, counts cost and applies budgets. See Setting up the AI gateway.

AI judgments (Jev). An AI model that StackJack asks to judge the text it scans. It can add a finding or raise a severity. It never removes a finding. See AI judgments (Jev).

Alert. A message that StackJack raises when it notices something a person should hear about. Every alert is in the Alerts inbox. A channel can also send it elsewhere. See Alerts and the inbox.

Alert channel. A place alerts are sent: email, Microsoft Teams, Slack, a PSA ticket, a signed webhook, or a SIEM.

Alert policy. A rule about which alerts go to which channels, and when. This is different from the AI use policy.

AI use policy. The policy your organization publishes for staff, built from a template. Staff accept each version. See The AI use policy and acknowledgements.

Audit trail. The record of who did what in AI Governance. It cannot be edited. See The audit trail.

Capture level. How much text StackJack keeps from a source: Facts only, Scan only, Masked text or Full text. See Capture levels and what your employer sees.

Client names. The list that lets a scan say which of your clients a piece of AI use was about. Also called the client dictionary. See Client names.

Client organization. For an MSP: one of your clients, governed in its own AI Governance organization. See Client organizations.

Connection. A saved login to an outside product, kept on the Connectors page. Vendor sources read through a connection.

Data-loss rule. A rule that looks for something in the text of an AI use. Also called a rule.

Data request. A request from a person about their data: an export or an erasure. See What staff can see about themselves.

Fact. One record of one AI use: who, when, which tool and model, tokens, cost and whether a rule matched.

Finding. The record that a rule matched. It shows the rule, not the text. See Findings and triage.

Framework. A set of requirements that auditors or insurers ask about, such as the EU AI Act or SOC 2. See Compliance readiness and the frameworks.

Incident. A record that someone opened for something that happened, with an owner, a timeline and a review. See Incidents and notification clocks.

Ingest token. The secret a telemetry sender uses to send data to one source. StackJack shows it once.

Inventory. The list of AI tools, models, keys, extensions, MCP servers and agents that StackJack has seen. See The AI tool inventory.

Jurisdiction. Where your staff are. It sets the default name policy, keep times and whether collection waits for a staff notice.

Legal hold. A hold that stops the keep times from deleting what a case may need. Business and above.

Masked token. A placeholder that stands in for a matched value. A finding shows the masked token, never the value.

MCP proxy. A StackJack address between your staff's AI tools and a remote MCP server you approved. See Setting up the MCP proxy.

Mode (of a rule). Off, Test or Live. A rule in Test mode keeps its findings and never alerts or changes stored text.

Name policy. Whether people who read governance results see staff names or pseudonyms.

Notification clock. A legal deadline on an incident, such as a 72-hour notice. StackJack records it and warns the owner.

Overview. The home page of AI Governance.

Personal key. A key that gives one person access to the AI gateway and the MCP proxy. See Your personal AI key.

Pseudonym. A label such as "Staff member 4F2A" that stands in for a name.

Pull source. A source that StackJack reads on a schedule from a vendor's admin account. Also called a vendor source.

Readiness. How far your evidence goes toward each framework. See Compliance readiness and the frameworks.

Report pack. A dated ZIP file with a PDF to read and CSV files with the rows behind it. See Report packs, templates and verification.

Reveal. Showing the name behind a pseudonym. A reviewer asks. A reveal approver decides. Both are recorded.

Rule pack. A ready-made set of rules for one kind of risk. See Rule packs, modes and custom rules.

Severity. How serious a finding is.

Source. A way AI use reaches StackJack. See Choose how to see AI use.

Staff notice. The record that you told your staff how their AI use is recorded.

Telemetry. Usage data that a tool sends in the OpenTelemetry format. Also called OTLP. See Send telemetry from coding tools.

Transparency notice. The notice that staff see about AI Governance, built from your settings.