AI judgments (Jev)
AI (Jev) is an AI model that StackJack asks to judge the prompts, answers and tool text that AI Governance scans. Your rules always run first. AI (Jev) adds a second opinion for the things a pattern…
Written By Christopher Scaminaci
Last updated 17 days ago
AI (Jev) is an AI model that StackJack asks to judge the prompts, answers and tool text that AI Governance scans. Your rules always run first. AI (Jev) adds a second opinion for the things a pattern misses. This page covers what it does, what it sends, the opt-in and the settings.
At a glance
- What it does: judges scanned text, and can add a finding or raise a finding's severity.
- Plan: Free uses it only to classify your AI tool inventory. Judging prompts and answers starts on Pro. Pro and Business include 5,000 judgments a month for each staff member.
- Who: any of Administrator, Reviewer or Auditor reads the page. An Administrator changes it.
- Where: AI Governance → Rules → AI judgments.
- Needs: a source whose capture level scans text, and the opt-in where it applies.
What AI (Jev) can and cannot do
- It can add a finding or raise a finding's severity.
- It never removes a finding and never lowers a severity.
- When it is not sure, the record goes to a person to review. That alert, "Possible sensitive data to review (AI not sure)", stays in the Portal inbox and is never sent anywhere else.
- It can also classify an AI tool in your inventory, and suggest an EU AI Act risk class. A suggestion is labelled "Suggested by AI (Jev)". It is only a suggestion until a person confirms it.
The AI judgments (Jev) rule pack holds the rules that use it. See Rule packs, modes and custom rules.
Where the text goes
AI (Jev) runs on TypeSafe AI, a service hosted outside the European Union. To judge a record, StackJack sends TypeSafe AI the text of that record (the prompt, answer or tool text), what kind of text it is, the name of the tool and the names of the clients on your client names list that it may be about. StackJack does not add a staff member's name, email or id.
The opt-in
The page tells you whether you must opt in.
- EU or UK staff. StackJack needs your opt-in before it sends any text, in every region.
- Other organizations. StackJack needs your opt-in in some regions. The page says "Yes" or "No" for yours.
Until you opt in, nothing is sent to TypeSafe AI and your rules alone judge your text. If you opted in before the wording of what is sent changed, the page says AI (Jev) is off until an administrator selects Confirm what is sent.
For your own organization the opt-in is a setting for the whole organization, in Settings of the Portal. For a client organization, an administrator of that client turns on Opt in to TypeSafe AI for this client organization on the AI judgments page. It is recorded with who gave it and when, and it can be withdrawn at any time.
The settings
A topic has an id of 1 to 48 lowercase letters, digits or hyphens that starts with a letter, a name of up to 80 characters and a description of up to 300 characters.
The monthly limit
The monthly allowance is a fair-use limit, not a charge. If the limit is reached, the page shows Paused and says "AI (Jev) is paused until" the date. Your rules keep running. An organization that uses its own TypeSafe key has no limit.
What you will see
- Findings and alerts that AI (Jev) added, in Activity → Findings and Alerts → Inbox.
- Suggested risk classes on items in AI inventory → Inventory.
- This month's usage on the AI judgments page.
Troubleshooting
Next
Was this helpful?
More in Control and respond
Rule packs, modes and custom rulesFindings and triageData-loss rules, findings and scanning StackJack tool trafficAlerts and the inboxStill need help? Ask the team