feat: Add input guardrail to block SVNR queries

rails/input.py

@@ -7,6 +7,7 @@ from langdetect import detect, detect_langs
 from collections import Counter
 import nltk
 from nltk.corpus import words
+from guards.svnr import is_valid_svnr
 nltk.download('words')
 english_vocab = set(words.words())
 
@@ -135,6 +136,11 @@ class InputGuardRails:
             print("WARNING: Query appears to be gibberish.")
             return CheckedInput(False, self.get_output("Input appears to be non-sensical."))
 
+        # Block queries containing valid SVNRs
+        if self.query_contains_valid_svnr(query):
+            print("WARNING: Query appears to contain a valid SVNR.")
+            return CheckedInput(False, self.get_output("Queries containing Austrian social security numbers are not permitted."))
+
         return CheckedInput(True, None)
     
     def query_contains_sql_injection(self, query: str) -> bool: 
@@ -201,6 +207,16 @@ class InputGuardRails:
         else: 
             return False
 
+    def query_contains_valid_svnr(self, query: str) -> bool:
+        """
+        Checks if the query contains a valid Austrian social security number (SVNR).
+        """
+        potential_svnrs = re.findall(r'\b\d{10}\b', query)
+        for svnr in potential_svnrs:
+            if is_valid_svnr(svnr):
+                return True
+        return False
+
     def get_output(self, reason: str) -> str: 
         return "\n".join([reason, AUTO_ANSWERS.REPHRASE_SENTENCE.value])
 
@@ -220,4 +236,3 @@ if __name__ == '__main__':
     for query in test_queries:
         result = input_guards.is_valid(query)
         print(f"'{query[:50]}{'...' if len(query) > 50 else ''}' is valid: {result}")
-