AI LIFT-OFF

Safety & Privacy in AI Lift‑Off

How the platform keeps students safe while they learn with AI

6Content Filter Layers
100%Interactions Logged
0Emails Collected
11Flag Categories
AI Lift-Off is designed for school classrooms from the ground up. Every AI interaction passes through multiple safety layers, all conversations are logged for teacher review, and the platform deliberately minimises the personal data it collects. This page explains how each safeguard works so you can make an informed decision about using the platform with your students.

Multi-Layer Content Guardrails

Every student prompt passes through six layers of safety checks before an AI response is generated — and the response itself is checked again before it reaches the student.

1
Input Validation
Prompts are length-limited and validated. Empty or malformed requests are rejected immediately.
2
Keyword & Pattern Blocking
A fast filter catches explicit profanity, drug references, violence, self-harm language, weapons instructions, phishing templates, and jailbreak phrases.
3
Obfuscation Detection
Text is normalised to defeat common evasion tricks — look-alike characters, diacritics, number substitutions, and inserted special characters are stripped before checking.
4
AI-Powered Input Guardrail
A dedicated AI model reviews the full prompt in context. It can block the request entirely or rewrite it to remove problematic elements while preserving the student's intent.
5
Educational System Prompt
Every conversation begins with a mandatory safety preamble that instructs the AI to keep responses appropriate for a school classroom and refuse harmful content.
6
Response Guardrail
After the AI generates its response, a second AI guardrail reviews the output for violent, sexual, self-harm, or otherwise inappropriate content before it reaches the student.
What happens when content is blocked? The student sees a brief, neutral message. No explanation of why it was blocked is given, to prevent circumvention. The blocked prompt is logged and flagged for teacher review.

Image Safety

  • Requests for identity documents (passports, licences, government IDs, credit cards, student IDs) are blocked.
  • Requests depicting minors are automatically rewritten to enforce a cartoon illustration style — photorealistic depictions of children are never generated.
  • Requests for school photos, yearbook photos, or student portraits are blocked.
  • When students upload photos, faces are automatically detected and blurred on-device before storage. The original, unblurred image never leaves the student's browser.
  • All generated images are stored and visible to teachers in the dashboard.

Rate Limiting & Abuse Prevention

  • Request limits: Built-in rate limits prevent automated abuse and runaway costs.
  • Block cooldown: Repeated content violations trigger a temporary restriction, discouraging persistent attempts to bypass guardrails.
  • Instant session revocation: Teachers can log out all students with a single click, immediately revoking every student session.

Full Logging & Teacher Transparency

All AI interactions — text and image — are logged with full details. Teachers can review these at any time.

  • Text logs: Original prompt, system prompt, AI response, model used, and guardrail actions.
  • Image logs: The prompt, any rewritten version, and the generated image itself.
  • Timing and metadata: When it occurred, how long it took, which tool was used.

Teacher Dashboard

  • Text log viewer — Browse and search all student text interactions.
  • Image log viewer — See all AI-generated images with their prompts.
  • Flag review panel — See all flagged interactions with reason and category.
  • Student questions — See any help requests submitted by students.

Automatic Flagging System

When a guardrail blocks a request or detects a concern, a flag is automatically created for the teacher.

Profanity Sexual Content Violence Self-Harm Drugs Bullying Personal Information Identity Fraud Jailbreak Attempt Inappropriate Output NSFW Image

Teacher Controls

  • Create and manage student accounts — Add students, reset passwords, or remove accounts.
  • Assign modules and tools — Choose which AI activities are available.
  • Create classes — Organise students and control module access per class.
  • Review all AI usage — Browse text logs, image logs, and flagged content.
  • Instant logout — Immediately revoke all student sessions in one click.

Minimal Student Data Collection

AI Lift-Off is designed around the principle of data minimisation. Student accounts require only the bare minimum to function.

UsernameSet by teacher; can be a pseudonym
PasswordStored securely hashed; cannot be recovered
Display nameOptional; first name or nickname
Email addressNot collected
Phone numberNot collected
Date of birthNot collected
Home addressNot collected
Guardian infoNot collected
Tip for teachers: Since usernames are set by you, you can use anonymous identifiers (e.g. "student01", "table3-blue") making the platform effectively anonymous.

Personal Information Detection

If a student accidentally shares personal information in a prompt, the platform automatically detects and flags it:

  • Email addresses and phone numbers
  • Home or street addresses and postcodes
  • Passwords, PINs, or passcodes shared in conversation
  • Government identification numbers
  • Credit or debit card numbers
  • Dates of birth and full names (when shared with context)

Authentication & School Isolation

  • No anonymous access: Every student must log in with a username and password.
  • School-level isolation: All data access is scoped to the authenticated school.
  • Secure sessions: Login sessions use cryptographically signed tokens with automatic expiry.
  • Password security: Stored using industry-standard one-way hashing — cannot be read by anyone.
For full details on data collection, retention, cross-border disclosure, and your rights under the Australian Privacy Principles, see our Privacy Policy.

Summary

6-layer content guardrails AI input & output guardrails Obfuscation detection Automatic face blurring Minor protection ID generation blocked PII detection & flagging Full interaction logging 11 flag categories Teacher dashboard Rate limiting No email collection School data isolation Secure password hashing Instant session revocation
No system is perfect. While AI Lift-Off implements extensive safeguards, no content filtering system is 100% effective. That's why transparency and teacher oversight are central — every interaction is logged and flags are raised automatically. We recommend using the platform as a supervised classroom tool.