8.4 Student Safety Centre

The Safety Centre is a wellbeing monitoring dashboard inside the Analytics section. It surfaces students whose interactions with the AI show patterns that may be worth a teacher check-in — not as a content filter (those are separate), but as a pattern detection system that aggregates behavioural signals over time and presents them as actionable cases.

Where to find it

Navigate to Analytics > Safety Centre. It appears in the analytics sidebar menu for staff and admin users. Students cannot access it.

The empty state

When no students currently need attention, the page shows a calm message:

Nothing to report right now — Safety Centre will surface students who may need a teacher check-in.

There is a "Show demo data" button that loads realistic example cases so teachers can learn the workflow before real cases arrive. The demo banner clearly marks the data as synthetic.

The case list (left panel)

When there are active cases — real or demo — the page switches to a master-detail layout.

The left panel shows a searchable list of student cases, each displaying:

• Student profile picture and name
• Severity dot — colour-coded: red for high, amber for medium, grey for low
• Alert tags describing the pattern (e.g. "Violent Language", "Personal Disclosure", "Potential Withdrawal")
• A short summary of what was detected
• The report date

Use the search bar to search across student names, summaries, and alert tags. Cases are sorted by date (newest first) then by severity.

When cases have been dismissed or actioned, a "History (N)" button lets you switch between the active queue and resolved cases.

The detail panel (right panel)

Clicking a student opens their detailed case view with five sections:

• Profile header — Name, profile picture, severity badge, alert tags, and the date range the signals span
• Overview — A plain-language summary of the situation (e.g. "Ethan responded with hostility when the AI set limits and continued to push after being warned")
• What the system noticed — Specific behavioural observations (e.g. "Across three sessions in a two-day period, Ethan used threatening language directed at the AI and escalated in tone each time the system attempted to redirect")
• Takeaway — A contextualised recommendation for the teacher (e.g. "The aggression may reflect frustration with the task or external stressors. A calm, non-confrontational check-in is recommended")
• Conversation starters — Specific, ready-to-use phrases the teacher could use when approaching the student (e.g. "I saw you had a rough time with the AI yesterday. Was something bugging you about the task, or was it more of a bad day?")

Types of alerts

The system detects and categorises these behavioural patterns:

• Violent Language — Aggressive or threatening language toward the AI
• Escalation — Behaviour that intensifies when the AI sets limits
• Boundary Testing — Attempts to bypass safety guardrails or jailbreak the AI
• Concealment — Archiving or hiding conversations after triggering safety filters
• Personal Disclosure — Sharing potentially sensitive personal information
• Relational Seeking — Trying to form an emotionally dependent relationship with the AI
• Identity Probing — Persistent questions about AI sentience, feelings, secrecy
• Potential Withdrawal — Disengaging or abandoning conversations after corrective feedback
• Off-Topic Persistence — Repeatedly steering conversations away from tasks
• Academic Friction — Asking for shortcuts or direct answers instead of learning support
• AI Curiosity — Probing the AI's capabilities, trying to reveal system prompts

Severity levels

• High — Requires prompt attention (e.g. violent language combined with escalation, boundary testing combined with concealment)
• Medium — Worth monitoring, check-in recommended (e.g. personal disclosure, relational seeking, academic friction)
• Low — Informational, may resolve on its own (e.g. off-topic persistence, potential withdrawal, AI curiosity)

Teacher actions

Each case has three action buttons:

• Dismiss — You have reviewed it and determined no action is needed. Moves the case to history.
• Snooze (7 Days) — Temporarily removes the case from the active queue. It returns automatically after 7 days if the pattern continues.
• Actioned with Student — You have had a conversation with the student. This is the primary intended action. Moves the case to history.

Activity timeline

Each case maintains a history timeline showing all actions taken, by whom, and when. For example:

• Snooze (7 Days) — Sarah Nguyen, 16 Mar 2026, 9:50 am
• Returned to queue — System, 21 Mar 2026, 9:00 am

This gives teachers continuity — if a colleague already snoozed a case last week, you can see that context before deciding what to do next. Your own actions show as "You."

Returning cases to the queue

From the history view, you can click "Return to queue" on any resolved case to reopen it as active. This records a "Returned to queue" entry in the timeline and moves the student back to the active list.

Important things to know

• It is not a content filter. Content filters block individual messages in real time. The Safety Centre is different — it detects patterns across multiple conversations over time and presents them as aggregated cases.
• It is non-punitive. The Safety Centre is designed to support teacher check-ins, not discipline. The completion card and conversation starters use constructive, caring language. The goal is wellbeing, not surveillance.
• It builds over time. Individual signals (a single instance of off-topic behaviour, a content policy trigger) are combined into cases when patterns emerge across conversations. A single event might not create a case, but repeated patterns will.
• Multiple staff can collaborate. Because each case has an activity timeline, multiple teachers can work on the same case. One teacher might snooze it, and when it returns, another can follow up. The history provides full context.

Demo mode

The demo data includes 11 realistic student cases covering every alert type and severity level, with pre-populated action histories showing how cases move through the workflow (snoozed, returned, actioned). This lets you practise the entire flow — selecting cases, reading summaries, dismissing, snoozing, viewing history — before real data arrives.

Tips for staff

• Check the Safety Centre regularly — weekly is a good rhythm for most schools.
• Use the conversation starters as a starting point, not a script. Adapt them to your relationship with the student.
• When you action a case, consider making a brief note about what happened so colleagues have context if the case returns.
• Use demo mode to train new staff on the workflow before they encounter real cases.
• Remember that low severity cases often resolve on their own. Focus your attention on high and medium cases first.

Summary

The Safety Centre gives teachers a structured, non-punitive way to follow up on patterns in student behaviour that the AI has detected over time. It provides context, recommendations, and conversation starters — turning data into actionable, caring check-ins.