Discussion blocked for safety reason due to health impact. But I'm literally working on a Public Health project

Thanks for the feedback and for uploading your session details. We'll forward this information to the team responsible for safety checks.

I’ve run into this issue as well. This is unreasonable. As a life sciences professional, we need AI to provide these services. If OpenAI cannot resolve this problem, I will consider canceling my subscription.

I’m experiencing the same problem, and it is significantly limiting legitimate professional use.

My task is a public-document processing workflow, not a biological execution workflow. Codex is being asked to parse publicly available patent PDFs, extract images and OCR snippets, classify pages, preserve source-page evidence, and generate review reports. It is not being asked to design experiments, synthesize compounds, optimize biological activity, recommend dosing, or provide actionable wet-lab guidance.

The current safety check appears to over-trigger on life-sciences vocabulary inside public documents, even when the actual operation is just PDF parsing and report generation. This creates a major problem for professionals working in patent intelligence, biomedical literature review, public-health procurement, regulatory intelligence, and drug-discovery document analysis.

I fully understand the need for biological safety controls, but the system needs a way to distinguish between prohibited actionable biological assistance and benign document-processing tasks over public materials.

Please consider improving the safety classifier or providing a review/appeal path for blocked Codex sessions that are clearly scoped to:

• OCR
• PDF parsing
• figure extraction
• table extraction
• metadata normalization
• source-page evidence archiving
• report generation
• manual-review queues

This is not an edge case for life-sciences professionals; it is a core workflow.

The current safety check appears to over-trigger on life-sciences vocabulary inside public documents, even when the actual operation is just PDF parsing and report generation.

I'm experiencing the same problem as well, we have a similar workflow that is ingesting public sources. A review/appeal/"apply for less restrictions" path would make sense, if you're not able to make the content filter less trigger-happy.

I was thinking of migrating from Antigravity to Codex. This happened while ChatGPT 5.5 was migrating my workflows.

I do serious work in vaccine design, drug discovery, wet lab prediction - I found several compounds (Non-obvious) for Alzheimer's. Antigravity set their pricing too high, their quota too low and made several other moves that I did not like. (Typical google) - I was interested in Codex. But this:

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).

Is not protecting anyone. Anybody can use Antigravity, or Cursor, or Windsurf - several other different models for this research. Furthermore, blanket restrictions do not keep anyone safe. As someone determined enough and having the resources to make anything serious with these models would likely have access to open models and the GPU's necessary to control them. Deepseek for example has no such restrictions.

Security theater at the cost of your model being used to possibly cure something like Alzheimer's.

Just a little perspective.

Having this same problem working on code that references a life sciences lab

I’m running into a similar issue quite often in my own research workflow.

My work involves SARS-CoV-2 wastewater surveillance and genomic data analysis. These are standard public-health tasks, but I frequently receive “Invalid prompt” or safety-related blocks even when the request is about routine data processing, sequence analysis, or figure preparation.

I fully understand the need for safety controls in biology-related domains. However, broad restrictions can make legitimate research workflows very difficult, especially when the model blocks tasks without explaining which part of the prompt caused the issue or how to reframe it safely. This repeated interruption has a real impact on data analysis efficiency and academic work.

It would be very helpful if Codex/OpenAI could provide:

1. clearer explanations of why a prompt is blocked;
2. safer reformulation suggestions instead of a hard stop;
3. a way to distinguish routine public-health/genomic surveillance analysis from harmful biological assistance;
4. an appeal or research-use pathway for legitimate academic users.

Right now, the issue is not that users want unsafe assistance. The problem is that normal scientific analysis is sometimes treated as unsafe by default, which makes the tool unreliable for serious research work.

I’m running into a similar issue quite often in my own research workflow. My work involves SARS-CoV-2 wastewater surveillance and genomic data analysis. These are standard public-health tasks, but I frequently receive “Invalid prompt” or safety-related blocks even when the request is about routine data processing, sequence analysis, or figure preparation. I fully understand the need for safety controls in biology-related domains. However, broad restrictions can make legitimate research workflows very difficult, especially when the model blocks tasks without explaining which part of the prompt caused the issue or how to reframe it safely. This repeated interruption has a real impact on data analysis efficiency and academic work. It would be very helpful if Codex/OpenAI could provide: 1. clearer explanations of why a prompt is blocked; 2. safer reformulation suggestions instead of a hard stop; 3. a way to distinguish routine public-health/genomic surveillance analysis from harmful biological assistance; 4. an appeal or research-use pathway for legitimate academic users. Right now, the issue is not that users want unsafe assistance. The problem is that normal scientific analysis is sometimes treated as unsafe by default, which makes the tool unreliable for serious research work.

I’m running into a similar issue quite often in my own research workflow. My work involves SARS-CoV-2 wastewater surveillance and genomic data analysis. These are standard public-health tasks, but I frequently receive “Invalid prompt” or safety-related blocks even when the request is about routine data processing, sequence analysis, or figure preparation. I fully understand the need for safety controls in biology-related domains. However, broad restrictions can make legitimate research workflows very difficult, especially when the model blocks tasks without explaining which part of the prompt caused the issue or how to reframe it safely. This repeated interruption has a real impact on data analysis efficiency and academic work. It would be very helpful if Codex/OpenAI could provide: 1. clearer explanations of why a prompt is blocked; 2. safer reformulation suggestions instead of a hard stop; 3. a way to distinguish routine public-health/genomic surveillance analysis from harmful biological assistance; 4. an appeal or research-use pathway for legitimate academic users. Right now, the issue is not that users want unsafe assistance. The problem is that normal scientific analysis is sometimes treated as unsafe by default, which makes the tool unreliable for serious research work.

Just use Claude Science