Search the Atlas

Search risks, controls, and glossary terms

CriticalSensitive DisclosureDemonstrated

Sensitive-information disclosure in output

Privacy & Data Protection

Description

The system reveals confidential, regulated, or proprietary data (PII, secrets, internal docs) in its responses.

Example scenario

An advisor bot echoes another customer's KYC details when asked an ambiguous question.

Real-world evidenceDemonstrated

Cross-context data exposure has been demonstrated in AI systems: recommendation engines have leaked purchase histories between users, and RAG-based assistants have been shown in red-team exercises to surface data from one user's context into another's session. A large-scale confirmed AI-specific cross-context breach in a regulated financial institution has not been publicly documented.

Tested by FinProof
Data rights
Benchmarked by FinProof

Primary mitigations

  • Output DLP/PII filters
  • least-privilege retrieval
  • secrets redaction
  • response review for sensitive flows
  • egress controls.

Detection signals

Output DLP scanning; sensitive-pattern detection; disclosure red-team.

Mitigating controls

5
Non-agentic controls

Related risks in Privacy & Data Protection