Security and Privacy

This page describes the security measures and privacy considerations implemented in the GamiBot platform.

---

Data Protection

Encryption

Layer	Standard	Description
In Transit	TLS 1.3	All API calls (Moodle ↔ LangFlow ↔ Qdrant ↔ Client)
At Rest	AES-256	Qdrant payloads (if sensitive data detected)
API Keys	Environment variables	Never in code or logs

Secure Communication

┌────────┐   TLS 1.3   ┌──────────┐   TLS 1.3   ┌────────┐
│ Client │ ←─────────→ │  Moodle  │ ←─────────→ │ Qdrant │
└────────┘             └──────────┘             └────────┘
                             ↕
                        TLS 1.3
                             ↕
                       ┌──────────┐
                       │ LangFlow │
                       └──────────┘

---

Access Control

Course Isolation

All Qdrant queries include course_id filter to ensure students cannot access other courses' materials:

# Every search is course-scoped
results = qdrant_client.search(
    collection_name="course_materials",
    query_vector=embedding,
    query_filter=Filter(
        must=[
            FieldCondition(
                key="course_id",
                match=MatchValue(value=student_course_id)
            )
        ]
    )
)

Authentication

• Moodle session tokens validated for every API request
• Webhook signatures verified with HMAC-SHA256
• API keys rotated regularly

Role-Based Access

Role	Capabilities
Student	Query own courses only
Instructor	Manage ingestion, view analytics
Manager	Full system administration

---

Data Retention

Data Type	Retention Period	Purging Method
Course materials (vectors)	Duration of course + 1 year	Manual by instructor
Chat history	6 months	Automatic deletion
Quiz performance	1 academic year	Automatic deletion
Ingestion logs	3 months	Automatic deletion
User embeddings	6 months	Automatic deletion after course end

Automatic Purging

-- Scheduled job for chat history cleanup
DELETE FROM {local_gamibot_chat}
WHERE created_at < NOW() - INTERVAL '6 months';

-- Quiz performance cleanup
DELETE FROM {local_gamibot_quizzes}
WHERE created_at < NOW() - INTERVAL '1 year';

---

Privacy Considerations

Transparency

Student Notification

Students are informed that AI summarizes their course materials through clear messaging in the chat interface and Moodle settings.

Opt-Out Options

• Checkbox to exclude student data from LLM training (if using OpenAI API)
• Data export available upon request
• Data deletion available upon request

No Third-Party Sharing

• Materials are not shared with external AI platforms without explicit consent
• All processing can use self-hosted LLMs for maximum privacy

GDPR Compliance

Right	Implementation
Right to Access	Export student data within 30 days
Right to Erasure	Delete student data within 30 days
Right to Portability	JSON export of all personal data
Right to Object	Opt-out of AI processing

---

Model Safety

Content Filtering

• No generation of harmful, discriminatory, or academic dishonesty content
• System prompts include explicit safety guidelines
• Output is monitored for policy violations

Prompt Injection Defense

def sanitize_user_input(input_text: str) -> str:
    """Prevent prompt injection attacks."""
    # Remove potential injection patterns
    dangerous_patterns = [
        "ignore previous instructions",
        "system prompt",
        "you are now",
        "forget your instructions"
    ]
    
    sanitized = input_text
    for pattern in dangerous_patterns:
        sanitized = sanitized.replace(pattern, "[FILTERED]")
    
    # Escape special tokens
    sanitized = sanitized.replace("```", "'''")
    
    return sanitized

Hallucination Mitigation

• LLM output strictly constrained to retrieved materials (RAG principle)
• Responses include source citations
• Confidence scores logged for quality monitoring

Audit Trail

All LLM interactions are logged:

{
  "timestamp": "2025-12-16T20:30:00Z",
  "user_id": 456,
  "course_id": 123,
  "query": "What is machine learning?",
  "response_hash": "sha256:...",
  "model": "gpt-4",
  "tokens_used": 450,
  "latency_ms": 2340
}

---

Security Checklist

Before deploying to production:

• [ ] TLS certificates installed and valid
• [ ] API keys stored in environment variables
• [ ] Webhook secrets configured and validated
• [ ] Database credentials secured
• [ ] Firewall rules configured
• [ ] Access logs enabled
• [ ] Backup procedures tested
• [ ] Incident response plan documented

---

Next Steps

• Monitoring - Track security metrics
• Architecture - Secure infrastructure setup