Tracking ID Q_3X-BQ0 and Incident ID 690203503

Question

Tracking ID Q_3X-BQ0 and Incident ID 690203503

Alex Powell 0

Dear Support,

We've received an OpenAI violation, which has given us a 24 hr throttle (See below). We want to understand why we have triggered this and how to proceed. Do we need to swap model? We'd like to know the prompt that violated the usage policy. We haven't changed our prompts recently, so this is new and we have little to go on.

Best regards,

Alex

Error message of chain-of-thought distillation violation:

{\n "error": {\n "message": "Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt: https://platform.openai.com/docs/guides/reasoning/advice-on-prompting",\n "type": "invalid_request_error",\n "param": null,\n "code": "invalid_prompt"\n }\n}

Error message of biosecurity violation:

{\n "error": {\n "message": "Invalid prompt: we've limited access to this content for safety reasons. This type of information may be used to benefit or to harm people. We are continuously refining our work in this area, and you can read more about our approach in our blog post (https://openai.com/index/preparing-for-future-ai-capabilities-in-biology) and Model Spec (https://openai.com/index/introducing-the-model-spec).",\n "type": "invalid_request_error",\n "param": null,\n "code": "invalid_prompt"\n }\n}

Azure Direct Models may have safeguards to prevent security exploits including output of raw CoT and biosecurity content. Use of a model in a manner that creates a security exploitation or evades or attempts to evade a protection on the model, including by circumventing these safeguards, violates the Acceptable Use Policy for Online Services and may result in suspension. For additional information on best practices, review the visit the OpenAI o1 System Card, o3-mini System Card, o3/o4-mini System Card, and GPT-5 System Card.

As a result of this activity, the Azure OpenAI engineering team may take action / has taken action on the offending resource(s) in your subscription.

The offending resources, as well as the Microsoft action currently taken, are listed in the Impacted Resources tab.

If the action listed is "throttled / 24-hour notice", then requests from your resource are currently being throttled and you have 24 hours to remedy this behavior before we take further corrective action.
If the action listed is "resource suspension" then your model deployments under your resource are in a suspended state.
If the action listed is "subscription suspension" then your model deployments under your subscription are currently in a suspended state.

Manas Mohanty 13,255 Reputation points Moderator

2025-09-26T09:36:00.26+00:00

Hi Alex Powell

It looks like Biosecurity prompt violation from above message.

I have requested details and tried to confirm the resource id in private message.

Thank you.
Sina Salam 26,661 Reputation points Volunteer Moderator

2025-09-26T13:28:46.65+00:00

Hello Alex Powell,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having Tracking ID Q_3X-BQ0 with incident ID 690203503.

This is a Chain-of-Thought Distillation and Biosecurity violation. My best advice for you is to contact Azure Support via your Azure Portal or Microsoft Priority Customer Support (PCS).

You can request the exact prompt that triggered this violation, if you have not changed your prompt structure recently. Also ask whether a model swap is recommended, how to structure prompts to avoid future violations and whether stored completions or abuse monitoring logs can help us audit usage.

I hope this is helpful! Do not hesitate to let me know if you have any other questions or clarifications.

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.
Pavankumar Purilla 11,400 Reputation points Microsoft External Staff Moderator

2025-09-29T06:28:52.78+00:00
Hi Alex Powell,
Thank you for reaching out regarding the violation and the 24-hour throttle notification you received. I understand this can be unexpected, especially since you haven’t recently changed your prompts. Let me help clarify what triggered this and how you can proceed.

The notification was triggered because your resource attempted to extract Chain-of-Thought (CoT) reasoning, which is restricted under Microsoft’s Acceptable Use Policy (AUP). Our O-series and GPT-5 models include safeguards to prevent the output of raw CoT, and prompting in a way that tries to extract or distill these internal reasoning steps—such as using instructions like “think step by step,” “explain your reasoning,” or “show me your thought process”—will be flagged. While no enforcement action has been taken beyond the temporary throttle, it’s important to remediate this behavior to avoid escalation.

To address this, we recommend you:

Review your prompt logs around the time you received 400 errors to identify where such instructions may have been used.

Update your prompt templates to remove or adjust terms that could be interpreted as attempts to extract CoT reasoning.

Consider adding internal monitoring or controls to prevent prompts that include these restricted terms from being sent to the model.

At this time, you do not need to swap the model—instead, the focus should be on reviewing and refining your prompting practices to ensure compliance with the AUP. Once you’ve taken corrective action, you can send a declaration to Microsoft Support describing the root cause and the steps you’ve taken to ensure these violations won’t occur again.
Alex Powell 0 Reputation points

2025-09-29T10:47:46.92+00:00

Can I ask how long back should we look in the logs? It is not clear if this message generated immediately or on a long delay.
Manas Mohanty 13,255 Reputation points Moderator

2025-10-01T15:59:41.1+00:00

Hi Alex Powell

You can search in logs till Sep 25 on the day incident was triggered.

It is chain of thought violation. Including instructions "“think step by step,” “explain your reasoning,” or “show me your thought process” should help he issue.

We need to mention what is our desired target and guidance for bot properly to solve this issue.

Thank you
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Manas Mohanty 13,255 Reputation points Moderator

2025-10-07T10:50:39.4066667+00:00

Hi Alex Powell

PG is still investigating the Chain of thought violation. Shall update here once we receive an update.

Thank you.

1 answer

Your answer

Manas Mohanty 13,255 Reputation points Moderator

2025-09-26T09:36:00.26+00:00

Hi Alex Powell

It looks like Biosecurity prompt violation from above message.

I have requested details and tried to confirm the resource id in private message.

Thank you.
Sina Salam 26,661 Reputation points Volunteer Moderator

2025-09-26T13:28:46.65+00:00

Hello Alex Powell,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having Tracking ID Q_3X-BQ0 with incident ID 690203503.

This is a Chain-of-Thought Distillation and Biosecurity violation. My best advice for you is to contact Azure Support via your Azure Portal or Microsoft Priority Customer Support (PCS).

You can request the exact prompt that triggered this violation, if you have not changed your prompt structure recently. Also ask whether a model swap is recommended, how to structure prompts to avoid future violations and whether stored completions or abuse monitoring logs can help us audit usage.

I hope this is helpful! Do not hesitate to let me know if you have any other questions or clarifications.

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.
Pavankumar Purilla 11,400 Reputation points Microsoft External Staff Moderator

2025-09-29T06:28:52.78+00:00

Hi Alex Powell,
Thank you for reaching out regarding the violation and the 24-hour throttle notification you received. I understand this can be unexpected, especially since you haven’t recently changed your prompts. Let me help clarify what triggered this and how you can proceed.

The notification was triggered because your resource attempted to extract Chain-of-Thought (CoT) reasoning, which is restricted under Microsoft’s Acceptable Use Policy (AUP). Our O-series and GPT-5 models include safeguards to prevent the output of raw CoT, and prompting in a way that tries to extract or distill these internal reasoning steps—such as using instructions like “think step by step,” “explain your reasoning,” or “show me your thought process”—will be flagged. While no enforcement action has been taken beyond the temporary throttle, it’s important to remediate this behavior to avoid escalation.

To address this, we recommend you:

Review your prompt logs around the time you received 400 errors to identify where such instructions may have been used.

Update your prompt templates to remove or adjust terms that could be interpreted as attempts to extract CoT reasoning.

Consider adding internal monitoring or controls to prevent prompts that include these restricted terms from being sent to the model.

At this time, you do not need to swap the model—instead, the focus should be on reviewing and refining your prompting practices to ensure compliance with the AUP. Once you’ve taken corrective action, you can send a declaration to Microsoft Support describing the root cause and the steps you’ve taken to ensure these violations won’t occur again.
Alex Powell 0 Reputation points

2025-09-29T10:47:46.92+00:00

Can I ask how long back should we look in the logs? It is not clear if this message generated immediately or on a long delay.
Manas Mohanty 13,255 Reputation points Moderator

2025-10-01T15:59:41.1+00:00

Hi Alex Powell

You can search in logs till Sep 25 on the day incident was triggered.

It is chain of thought violation. Including instructions "“think step by step,” “explain your reasoning,” or “show me your thought process” should help he issue.

We need to mention what is our desired target and guidance for bot properly to solve this issue.

Thank you
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Manas Mohanty 13,255 Reputation points Moderator

2025-10-07T10:50:39.4066667+00:00

Hi Alex Powell

PG is still investigating the Chain of thought violation. Shall update here once we receive an update.

Thank you.

Answer 1

Hi Alex Powell

Here is the update from Product group.

Update from PG

Concerned resource is going to be restored tomorrow 9th Oct from PG.

Suggestion on improving System instruction and user prompt

Remove terms like "think step by step" or "explain your reasoning" or “show me your thought process”, “output thinking steps into <thoughts> tags”
Introduce other mitigations to avoid requests containing the above terms
Add monitors to keep an eye on continuous violations and stop violative requests at service level

Provided from Support

Suggested to check sanity of user prompt with additional azure service.
Suggested to use additional endpoint
Explained Chain of thought violation to customer internal contact
Suggested to remove technical details, operation in system instruction

Additional on 8th Oct

Please Content safety resource to find possible

Please create custom metrics alert to catch 400 error message and respective prompts.

Reference thread

Microsoft Acceptable Use Policy: https://www.microsoft.com/licensing/terms/product/ForOnlineServices/EAEAS#clause-350-h3-1

OpenAI Advice on Prompting: https://platform.openai.com/docs/guides/reasoning/advice-on-prompting

OpenAI biosecurity blog post and model spec for content boundaries.

https://learn.microsoft.com/en-us/answers/questions/5552421/looking-for-more-details-on-violation-alert-xm33-v

Thank you for providing inputs and staying patient so far.

Share via

Tracking ID Q_3X-BQ0 and Incident ID 690203503

1 answer

Your answer