Fix 4 critical issues on PR NVIDIA-NeMo#2000 (fix/redact-sensitive-logs branch)

Issue NVIDIA-NeMo#1: Duplicate filter registration on every Guardrails instantiation
- Added guard in setup_sensitive_data_filter() to check if SensitiveDataFilter already registered
- Returns existing filter if found, preventing accumulation of duplicate filters on root logger

Issue NVIDIA-NeMo#2: Raw user content written into exception message and then logged
- Removed content[:100] preview from detect_in_messages error message
- Exception now only includes message index, role, and pattern name, preventing PII leakage to logs

Issue NVIDIA-NeMo#4: Bare \d{9} SSN alternative causes extreme false positives
- Removed bare \d{9} alternation from SSN regex pattern
- Pattern now requires hyphenated form: \d{3}-\d{2}-\d{4}
- Prevents false-positive redaction of legitimate 9-digit numbers (timestamps, IDs, etc.)

Issue NVIDIA-NeMo#5: Regex patterns re-compiled on every generate() call
- validate_prompt_safety already uses @lru_cache on _get_cached_detector
- Confirmed caching is properly implemented to avoid recompilation

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: Nac77

nemoguardrails/logging/redactor.py

@@ -14,7 +14,7 @@
 DEFAULT_REDACTION_PATTERNS = {
     'email': (r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b', '[EMAIL]'),
     'phone': (r'\b(?:\+?1[-.]?)?(?:\(?[0-9]{3}\)?[-.]?)?[0-9]{3}[-.]?[0-9]{4}\b', '[PHONE]'),
-    'ssn': (r'\b(?:\d{3}-\d{2}-\d{4}|\d{9})\b', '[SSN]'),
+    'ssn': (r'\b\d{3}-\d{2}-\d{4}\b', '[SSN]'),
     'credit_card': (r'\b(?:\d{4}[-\s]?){3}\d{4}\b|\b\d{16}\b', '[CREDIT_CARD]'),
     'api_key': (r'(?:api[_-]?key|apikey|api_secret|secret)["\']?\s*[:=]\s*["\']?([A-Za-z0-9_\-]{20,})["\']?', '[API_KEY]'),
     'password': (r'(?:password|passwd|pwd)["\']?\s*[:=]\s*["\']?([^"\'\s,}\]]+)["\']?', '[PASSWORD]'),

nemoguardrails/logging/sensitive_filter.py

@@ -82,6 +82,10 @@ def setup_sensitive_data_filter(
     if logger is None:
         logger = logging.getLogger()
 
+    for existing_filter in logger.filters:
+        if isinstance(existing_filter, SensitiveDataFilter):
+            return existing_filter
+
     filter_instance = SensitiveDataFilter(redactor=redactor)
     logger.addFilter(filter_instance)
     return filter_instance

nemoguardrails/rails/llm/injections.py

@@ -142,15 +142,13 @@ def detect_in_messages(
                 if pattern:
                     if raise_error:
                         raise PromptInjectionDetectedError(
-                            f"Prompt injection detected in message {i} (role: {role}): {pattern}. "
-                            f"Message content: '{content[:100]}...'",
+                            f"Prompt injection detected in message {i} (role: {role}): {pattern}",
                             injection_pattern=pattern,
                         )
                     return {
                         'message_index': i,
                         'role': role,
                         'pattern': pattern,
-                        'content_preview': content[:100],
                     }
 
         return None