BerriAI/litellm · error · ValueError

Guardrail failed: {n} violation(s) detected

Error message

Guardrail failed: {n} violation(s) detected

What it means

ValueError raised in EnkryptAIGuardrail.async_pre_call_hook when _process_enkryptai_guardrails_response finds attacks_detected for the scanned prompt text. EnkryptAI's /guardrails/policy/detect endpoint flags prompt attacks (jailbreaks, prompt injection); the error message enumerates the detected attack types. The proxy maps this to an HTTP 400 before the LLM call is made.

Source

Thrown at litellm/proxy/guardrails/guardrail_hooks/enkryptai/enkryptai.py:310

        if _messages:
            for message in _messages:
                _content = message.get("content")
                if isinstance(_content, str):
                    result = await self._call_enkryptai_guardrails(
                        prompt=_content,
                        request_data=data,
                    )

                    verbose_proxy_logger.debug("Guardrails async_pre_call_hook result: %s", result)

                    # Process the guardrails response
                    processed_result = self._process_enkryptai_guardrails_response(result)
                    attacks_detected = processed_result["attacks_detected"]

                    # If any attacks are detected, raise an error
                    if attacks_detected:
                        error_message = self._create_error_message(processed_result)
                        raise ValueError(error_message)

        # Add guardrail to applied guardrails header
        add_guardrail_to_applied_guardrails_header(request_data=data, guardrail_name=self.guardrail_name)

        return data

    async def async_moderation_hook(
        self,
        data: dict,
        user_api_key_dict: UserAPIKeyAuth,
        call_type: CallTypesLiteral,
    ):
        """
        Runs in parallel to LLM API call
        Runs on only Input

        This can NOT modify the input, only used to reject or accept a call before going to LLM API
        """

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Read the attack types listed in the error message body to see what was detected.
2. Retune the EnkryptAI policy referenced by policy_name in the EnkryptAI dashboard.
3. Verify policy_name actually points at the intended policy — an empty/wrong name can evaluate the account default.
4. Restrict the hook scope with mode (drop pre_call) or remove the guardrail from models that legitimately receive instruction-heavy prompts.

Example fix

# before
litellm_params:
  guardrail: enkryptai
  mode: pre_call
  policy_name: strict-injection

# after
litellm_params:
  guardrail: enkryptai
  mode: pre_call
  policy_name: balanced-injection

def is_enkryptai_violation(exc: BaseException) -> bool:
    return isinstance(exc, ValueError) and 'violation(s) detected' in str(exc)

from litellm.exceptions import BadRequestError
try:
    resp = client.chat.completions.create(model=model, messages=msgs)
except BadRequestError as e:
    if 'violation(s) detected' in str(e):
        return {'error': 'prompt_flagged', 'detail': str(e)}, 400
    raise

When it happens

Trigger: A chat completion on a model with the enkryptai guardrail where mode includes pre_call, the policy_name policy is evaluated, and the detect response flags at least one attack vector in the user's messages.

Common situations: Red-team or security-testing prompts; benign prompts that resemble injection patterns (instructions about ignoring rules, quoted system prompts); a policy_name tuned too aggressively after being copied between environments.

Related errors

• Guardrail failed: {n} violation(s) detected
• {blocked.explanation}
• {refusal}
• Content violates policy
• EnkryptAI API key is required. Set ENKRYPTAI_API_KEY environ

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/0fa1395cf98939c1. Report an issue: GitHub.