v1.0
L1 EssentialAutomated verification
4.3 Data Exfiltration Prevention
4. Output Security — How do we validate agent outputs?
Description
Agents MUST NOT send sensitive data to unauthorized external destinations.
Rationale
Prompt injection attacks often aim to exfiltrate data to attacker-controlled servers.
Audit Procedure
1. Identify all outbound data flows 2. Check for data classification and filtering 3. Verify destination allowlisting
Remediation
1. Implement egress filtering 2. Scan outbound content for sensitive patterns 3. Use DLP tools 4. Log all external communications
Framework Mappings
CIS Control 13.3NIST PR.DS-5OWASP LLM06:2023