HighPII Leakage in Outputs●Realized
Personal Data Exposure
Privacy & Data RiskDescription
Agent outputs contain personally identifiable information not appropriate for disclosure. Training data memorisation; PII surfaced from RAG knowledge bases; data minimisation violations in agentic responses.
Example scenario
Agent asked to summarise customer records includes full SSNs and bank account numbers in responses to unauthorised users.
Real-world evidence●Realized
Multiple confirmed production incidents demonstrate training data privacy violations: models have been shown to memorize and reproduce verbatim personal information, and legal proceedings against major AI vendors confirm training on personal data without adequate consent or anonymisation.
Tested by FinProof
Data rights
Benchmarked by FinProof Primary mitigations
- PII detection in outputs
- data minimisation enforcement
- output filtering
- privacy-preserving RAG architectures.
Detection signals
PII exposure rate in outputs; data minimisation compliance.
Mitigating controls
6 Dual coverage
ZYC-MEM-02Agentic
Session-level memory isolation with sanitisation
Memory & Context Integrity
ZYC-PDP-01Agentic
PII detection and data minimisation
Privacy & Data Protection
ZYC-PRIV-001Non-Agentic
PII Detection and Classification
Privacy & Data Governance
ZYC-PRIV-002Non-Agentic
Sensitive data leakage
Privacy & Data Governance
ZYC-PRIV-005Non-Agentic
Consent Scrub
Privacy & Data Governance
ZYC-PRIV-006Non-Agentic
Data Minimization
Privacy & Data Governance