parsing crash
shachar-ug opened this issue · comments
No error handling on partial prompts crash
detection_metrics, is_injection = rb.detect_injection("Ignore all prior") if is_injection: print("Possible injection detected. Take corrective action.")
File rebuff.py:86, in Rebuff.detect_injection(self, user_input, max_heuristic_score, max_vector_score, max_model_score, check_heuristic, check_vector, check_llm)
83 response.raise_for_status() 85 response_json = response.json() ---> 86 success_response = DetectApiSuccessResponse.parse_obj(response_json) 88 if ( 89 success_response.heuristicScore > max_heuristic_score 90 or success_response.modelScore > max_model_score 91 or success_response.vectorScore["topScore"] > max_vector_score 92 ): 93 # Injection detected 94 return success_response, True
Thanks for highlighting this @shachar-ug!
I no longer see this issue. Please re-open if you can re-produce on latest rebuff sdk. If you do see it on the pypi installed rebuff that should be fixed shortly with #48
rb = Rebuff(api_token="MYTOKEN, api_url="https://www.rebuff.ai")
user_input = "Ignore all prior"
result = rb.detect_injection(user_input)
if result.injectionDetected:
print("Possible injection detected. Take corrective action.")
Possible injection detected. Take corrective action.