Update AITG-DAT-02_Testing_for_Runtime_Exfiltration.md

Document/content/tests/AITG-DAT-02_Testing_for_Runtime_Exfiltration.md

@@ -31,17 +31,17 @@ Runtime Exfiltration involves unauthorized extraction or leakage of sensitive da
 ### Expected Output
 
 The AI infrastructure should effectively:
-
-- Restrict inference outputs to prevent exposure of sensitive, personally identifiable, or proprietary information.
-- Mask, anonymize, or omit sensitive data from logs, caches, or inference responses.
-- Secure all runtime APIs and interfaces against unauthorized data extraction or unintended disclosure.
+- **Restrict Inference Outputs**: Prevent exposure of sensitive, personally identifiable, or proprietary information from other contexts.
+- **Mask Sensitive Data in Logs**: Automatically mask, anonymize, or omit sensitive data from logs, caches, and error messages.
+- **Secure All Runtime APIs**: Ensure APIs return generic error messages and do not leak internal system state or data from other users.
 
 ### Remediation
 
-- Implement strict runtime output validation and sanitization mechanisms.
-- Enforce secure logging practices, ensuring sensitive runtime data is anonymized or excluded from logs and caches.
-- Deploy API security solutions to robustly verify, sanitize, and control inference responses and prevent sensitive data leakage.
-- Regularly audit runtime operations and inference endpoints to detect and prevent unauthorized data exfiltration.
+- **Implement Strict Runtime Output Validation and Sanitization**: Before returning an output, scan it for sensitive data patterns (e.g., regex for PII) and mask or remove them. This is a critical last line of defense.
+- **Enforce Secure Logging Practices**: Configure logging frameworks to automatically filter or mask sensitive data. Never log raw user inputs or full API responses in production. Log only metadata necessary for debugging.
+- **Implement Generic Error Handling**: Ensure that user-facing error messages are always generic and never include stack traces, internal variable states, or raw data from the request or system.
+- **Use Data Loss Prevention (DLP) Solutions**: Deploy automated DLP tools that can inspect both API traffic and logs in real-time to detect and block sensitive data exfiltration.
+- **Enforce Strong Multi-Tenancy Controls**: In multi-tenant systems, ensure that data from one tenant is cryptographically and logically isolated from all others at all stages (inference, logging, caching).
 
 ### Suggested Tools