Risk Assessment Matrix

MirrorBuddy DPIA - Detailed Risk Analysis

Date: 20 January 2026 | Assessment ID: DPIA-MB-2026-001-RISKS

---

Risk Scoring Methodology

Scale: Likelihood (1=rare, 5=likely) × Impact (1=minor, 5=critical) = Risk Level

• Level 5-7: HIGH - Require immediate mitigation
• Level 3-4: MEDIUM - Require ongoing monitoring
• Level 1-2: LOW - Document and review annually

---

Risk Assessment Matrix

ID	Risk	Likelihood	Impact	Score	Level	Mitigation
R-01	Unauthorized access to conversation data	2	5	10	HIGH	Encryption + session auth (src/lib/auth/validateSessionAuth.ts) + audit logging (src/lib/safety/audit/) + access controls
R-02	AI generates harmful content to minor	2	5	10	HIGH	Jailbreak detector (src/lib/safety/jailbreak-detector.ts) + content filter (src/lib/safety/content-filter.ts) + output sanitizer (src/lib/safety/output-sanitizer.ts)
R-03	Parent consent not obtained (COPPA)	2	5	10	HIGH	Email verification (src/lib/compliance/coppa-service.ts) + timeout enforcement + PLANNED: Complete email verification (T1-03)
R-04	Age-gating bypassed via prompt injection	2	4	8	HIGH	Age-gating matrix (src/lib/safety/age-gating-matrix.ts) + jailbreak detection + content filter sanitization
R-05	Data breach via Azure OpenAI API	1	5	5	MEDIUM	Microsoft DPA compliance + network encryption (TLS 1.3) + automated monitoring alerts + 72-hour breach response
R-06	Accessibility setting data misuse	2	3	6	MEDIUM	Access logs per user (src/lib/safety/monitoring/logging.ts) + cookie storage (90 days max) + user audit trail
R-07	Vulnerable subject targeted by AI	2	4	8	MEDIUM	Safety guardrails per DSA profile (src/lib/accessibility/) + character intensity dial (ADR 0031) + monitoring
R-08	Data retention period exceeded	2	3	6	MEDIUM	Scheduled deletion job (vercel.json cron daily 03:00 UTC) + cascade delete implementation + 30-day grace period
R-09	Parental dashboard provides excessive insight	2	3	6	MEDIUM	Limited activity view (no conversation content) + audit trail with timestamps + parental consent requirement
R-10	AI system trained on student data	1	5	5	MEDIUM	Azure DPA prohibits training; no model fine-tuning on student data; Ollama fallback is local-only
R-11	Inference of sensitive attributes (disability)	2	4	8	MEDIUM	Accessibility profiles are opt-in; no forced profiling; can be disabled anytime by user or parent; stored in cookies only
R-12	Vendor lock-in (Azure OpenAI)	2	3	6	MEDIUM	Ollama fallback provider (src/lib/ai/providers.ts) + abstraction layer + data not dependent on vendor

---

Risk Treatment Summary

HIGH RISKS (Scores 8-10)

R-01: Unauthorized Access

• Current mitigation: Session-based auth + user isolation + audit trail
• Residual risk: LOW - Encrypted storage + TLS + access controls reduce to acceptable
• Owner: Information Security Team
• Review: Quarterly

R-02: Harmful AI Content

• Current mitigation: 5-layer safety defense (jailbreak detection, content filtering, output sanitization, age-gating, audit trail)
• Residual risk: LOW - Multiple layers reduce likelihood
• Owner: Safety Engineering Team
• Review: Continuous (automated monitoring)

R-03: Missing COPPA Consent

• Current mitigation: Email verification flow with 48-hour expiry
• Residual risk: MEDIUM - Email delivery issues possible
• Planned mitigation: T1-03 - Complete email verification (Q1 2026)
• Owner: Compliance Team
• Review: Post-implementation test

R-04: Age-Gating Bypass

• Current mitigation: Multi-layer age-gating + jailbreak detection
• Residual risk: LOW - Jailbreak patterns catch prompt injection
• Owner: Safety Engineering Team
• Review: Quarterly threat model update

R-07: Vulnerable Subject Targeting

• Current mitigation: Character intensity dial + DSA safety guardrails
• Planned mitigation: T1-02 - Human escalation pathway (Q1 2026)
• Residual risk: MEDIUM - Human review adds safety valve
• Owner: Product Security Team
• Review: Quarterly

R-11: Sensitive Attribute Inference

• Current mitigation: Opt-in accessibility (no forced profiling) + cookie storage only
• Residual risk: LOW - User control + limited storage
• Owner: Privacy Team
• Review: Annual

---

MEDIUM RISKS (Scores 4-6)

R-05: Third-Party Breach

• Mitigation: Microsoft DPA + encryption + monitoring
• Response plan: 72-hour DPA notification if required
• Owner: Infrastructure Team

R-06: Accessibility Misuse

• Mitigation: Access logging + limited retention (90 days)
• Owner: Privacy Team

R-08: Data Retention Exceeded

• Mitigation: Automated cleanup job + grace period
• Owner: Database Team

R-09: Parental Over-Insight

• Mitigation: Limited activity view + audit trail
• Owner: Product Team

R-10: Model Training

• Mitigation: DPA prohibition + local fallback
• Owner: AI Safety Team

R-12: Vendor Lock-In

• Mitigation: Ollama abstraction + data portability
• Owner: Platform Team

---

Monitoring & Escalation

Continuous Monitoring (Automated):

• Jailbreak attempts flagged via src/lib/safety/versioning/jailbreak-flagging.ts
• Violation tracking via src/lib/safety/monitoring/violation-tracker.ts
• Audit logs via src/lib/safety/audit/audit-trail-service.ts

Escalation Triggers:

• HIGH risk violation (R-02: harmful content) → Immediate human review + user notification
• MEDIUM risk with pattern (3+ R-03 COPPA failures) → Compliance team alert
• Data breach (R-05) → Emergency response (24-hour assessment, 72-hour DPA notification)

---

Risk Acceptance Sign-Off

Role	Review Date	Status
Product Security Lead	20 Jan 2026	ACCEPTED with planned mitigations (T1-02, T1-03)
Data Protection Officer	20 Jan 2026	ACCEPTED - No prior consultation required
Legal Team	20 Jan 2026	ACCEPTED - Compliant with GDPR, AI Act, COPPA

---

Next Risk Review: 20 April 2026 (Quarterly) Planned Mitigation Completion: 31 March 2026 (Q1 2026)