The OpenAI API and Codex use additional safeguards for some requests involving cybersecurity or biological research and applications. For GPT-5.6, this additional-check experience applies only to certain models. The same experience will begin rolling out in ChatGPT when GPT-5.6 is available there. Depending on the request and recent activity on your account, you may see a message such as:
This request requires additional safety checks, which can take extra time. You can retry with a faster model, though it may be less capable of handling complex requests.
This content can’t be shown
Some models or app versions may use older wording, such as:
Your conversations have multiple flags for possible cybersecurity risk
Your conversations have multiple flags for possible biological safety risk
These notices come from automated safety systems. Seeing one does not, by itself, mean OpenAI has determined that you violated our Usage Policies.
What happens during an additional safety check?
In Codex, you’ll see This request requires additional safety checks, which can take extra time. You can retry with a faster model, though it may be less capable of handling complex requests. while an additional automated check runs. API requests may also take longer while a check is running. When 5.6 rolls out on ChatGPT, the same in-product message may also appear there. If the model’s response can be provided safely, it will continue after the check completes. If it cannot, you’ll see This content can’t be shown.
The check can make a response take longer than usual. You do not need to take any action while it runs.
Why am I seeing this message?
Additional checks can be triggered by:
The content of the current request or response.
A high volume of recent cybersecurity or biological requests that our automated systems identify as potentially risky.
When the second condition applies, additional checks may temporarily run on every request for your account when you are interacting with GPT-5.6 Sol and Terra—not only requests about cybersecurity or biology.
When GPT-5.6 is available in ChatGPT, these checks can apply at the account level across ChatGPT and Codex. This means recent activity in Codex may affect requests in ChatGPT, or vice versa, when you use the same OpenAI account.
Will the additional checks stop?
Account-level additional checks are temporary and depend on recent usage. Once recent activity no longer meets the applicable safety criteria and the temporary period ends, requests return to their usual behavior.
New potentially risky activity can extend that period, so there is no single duration that applies to every account.
Can I use a faster model?
If retry with a faster model appears, selecting it retries the same request with GPT-5.6 Luna, a faster but lower intelligence GPT-5.6 model that does not use this additional-check experience. The faster model may respond more quickly but may be less capable of handling complex tasks. Using a different model does not bypass OpenAI’s Usage Policies or other safeguards.
Please note that changing models is optional. Codex—and ChatGPT in the future—will not silently switch models through this control.
What can I do if content can’t be shown?
If content can’t be shown, review the request against OpenAI’s Usage Policies. If the request is allowed, try again with a narrower scope and only the context needed to explain the outcome you want:
For cybersecurity work, focus on a defensive outcome, such as identifying, preventing, or remediating a security issue. Omit exploit details that are not necessary to that outcome.
For biological work, focus on safety, prevention, analysis, or risk mitigation. Omit procedural details that are not necessary to that purpose.
Changing the wording does not change whether a request is allowed and does not guarantee that a response can be provided.
Can I apply for Trusted Access?
Trusted Access is intended to reduce unnecessary friction for eligible, approved work. Approval is not automatic, and other safeguards and access controls continue to apply.
Cybersecurity: Security professionals and organizations conducting authorized work can learn about eligibility and apply through Trusted Access for Cyber. For managed workspaces, Trusted Access is limited to approved internal users and workflows and cannot be extended to customers or third parties.
Biological research: Eligible researchers at approved organizations may be able to apply for Trusted Access. You can also read more about OpenAI’s approach to advanced biological capabilities.
What if I think the check or block is a mistake?
When available, you can use the thumbs-down control to share feedback about an individual response. This feedback helps improve our systems, but it does not immediately remove account-level checks.
If a clearly benign or authorized request is repeatedly blocked, contact OpenAI Support and include:
The exact safety message you received.
The model and whether you were using the OpenAI API, Codex, or ChatGPT.
The date, time, and time zone.
The request ID, if available.
A brief, redacted description of the task.
Your workspace or organization name or ID, if applicable.
Do not include passwords, authentication codes, proprietary data, or other sensitive information.