microsoft/semantic-kernel · error · ContentFilterAIException

{type(self)} service encountered a content error

Error message

{type(self)} service encountered a content error

What it means

Raised as ContentFilterAIException in _send_completion_request when the OpenAI SDK raises a BadRequestError whose code attribute equals 'content_filter'. This indicates the OpenAI content-moderation system blocked the prompt or completion. Semantic Kernel re-wraps it as a domain-specific ContentFilterAIException to distinguish it from other 400 errors.

Source

Thrown at python/semantic_kernel/connectors/ai/open_ai/services/open_ai_handler.py:96

        settings: OpenAIPromptExecutionSettings,
    ) -> ChatCompletion | Completion | AsyncStream[ChatCompletionChunk] | AsyncStream[Completion]:
        """Execute the appropriate call to OpenAI models."""
        try:
            settings_dict = settings.prepare_settings_dict()
            if self.ai_model_type == OpenAIModelTypes.CHAT:
                assert isinstance(settings, OpenAIChatPromptExecutionSettings)  # nosec
                self._handle_structured_output(settings, settings_dict)
                if settings.tools is None:
                    settings_dict.pop("parallel_tool_calls", None)
                response = await self.client.chat.completions.create(**settings_dict)
            else:
                response = await self.client.completions.create(**settings_dict)

            self.store_usage(response)
            return response
        except BadRequestError as ex:
            if ex.code == "content_filter":
                raise ContentFilterAIException(
                    f"{type(self)} service encountered a content error",
                    ex,
                ) from ex
            raise ServiceResponseException(
                f"{type(self)} service failed to complete the prompt",
                ex,
            ) from ex
        except Exception as ex:
            raise ServiceResponseException(
                f"{type(self)} service failed to complete the prompt",
                ex,
            ) from ex

    async def _send_embedding_request(self, settings: OpenAIEmbeddingPromptExecutionSettings) -> list[Any]:
        """Send a request to the OpenAI embeddings endpoint."""
        try:
            response = await self.client.embeddings.create(**settings.prepare_settings_dict())

View on GitHub (pinned to c028a0c7dc)

Solutions

  1. Sanitize or pre-filter user input before sending it to the model
  2. Add a retry with a rephrased or truncated prompt if the filter is a false positive
  3. Review OpenAI's moderation guidelines and adjust your application's content policy accordingly
  4. Consider using the OpenAI Moderation API to pre-screen input before the completion call

Example fix

# before
response = await service.get_chat_message_content(chat_history=history, settings=settings)
# after — pre-moderate then call
moderation = await client.moderations.create(input=user_text)
if moderation.results[0].flagged:
    return 'Content blocked by moderation policy.'
response = await service.get_chat_message_content(chat_history=history, settings=settings)
Defensive patterns

Strategy: try-catch

Validate before calling

from openai import AsyncOpenAI

async def input_is_safe(client: AsyncOpenAI, text: str) -> bool:
    result = await client.moderations.create(input=text)
    return not result.results[0].flagged

Try / catch

from semantic_kernel.connectors.ai.open_ai.exceptions.content_filter_ai_exception import ContentFilterAIException

try:
    response = await service.get_chat_message_content(...)
except ContentFilterAIException as e:
    logger.warning('Content filtered by OpenAI: %s', e)
    return safe_fallback_message

Prevention

When it happens

Trigger: Sending a prompt that triggers OpenAI's content filter — the input or generated output contains content that violates OpenAI's usage policies, causing the API to return a 400 with code 'content_filter'.

Common situations: Processing untrusted user input that contains prohibited content; aggressive system prompts; jailbreak attempts in production; testing with edge-case prompts that trip moderation.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/6083e7a228fcdcdd. Report an issue: GitHub.