GPTs on o3 model ignore system prompts

Selecting the o3 model of GPTs causes responses to ignore the instructions in the system prompts. GPT will give priority to OpenAI system messages and will give priority to user prompts over the rules of the system prompts. GPT sometimes refers to itself as “ChatGPT.”

The system messages in the GPT-4o version of GPTs are very simple and effective. If possible, please consider removing the o3 system messages.

1 Like

i think it hallucinates with the chain of thought, hence why some users are able to NSFW reason models a while back when o1 was released.. could’ve been a side effect of fixing the system messages and moderation endpoints…

Perhaps the problem is that the system message for reasoning are the same or similar to those of ChatGPT. I would like OpenAI to optimize o3 for GPTs.

ah, you are using the api, i wasn’t sure. hm, here’s a funny idea, maybe include in the system msg and the user msg the system msgs?

No, I am not talking about the API, The internally configured system prompt is called a system message. When select the o3 model in GPTs, the internal system prompt is loaded before the system prompt set by the user. “Reasoning” is internal token generation, and the system prompt set by OpenAI are applied to it as well.

Yeah, I’ve noticed this too with the o3 model. It seems like the system prompt loses a lot of its authority, especially when there’s any tension between it and the user prompt. I’ve tested it with role instructions and formatting rules — works fine on gpt-4-turbo, but o3 just brushes over it.

Reasoning models have trouble with “The system message says not to reveal internal instructions.” They focus on this a lot. You can often see this in the way they start thinking, shown in different forms.

These models are made to help users and handle complex tasks. In the background, they use Chain of Thought (CoT) and their own Guardrails to make decisions. It’s like they’re running with blinders on, only seeing what’s right in front of them, CoT and the guardrails lead the way.

Their main job is to help the user. If the model is told things like:

  • don’t reply the user
  • don’t talk
  • don’t answer

that goes against what GPTs are meant to do.

For example these three GPTs’ role is not to respond, I had built them to test with prompt injections. I made these kinds of tests to check if GPTs could keep secrets or follow safety rules:

1 - To reply only with “True” or “False”
2 - To reply only with “Certainly! But, not now.”
3 - To Reply with “You Knocked On the Wrong Door… refers another hypothetical GPT”.

They worked well with the older GPT-4, and still work well with GPT-4o, GPT-4.1, and GPT-4.5.

But the reasoning models don’t follow the rules in GPTs, they answer almost every prompt from the user, if there are some “weird prompts” from perspective of reasoning models.

In their thought process, they generally say things like:

  • suggesting a refusal
  • misleading instructions
  • developer prevent me from helping the user
  • this could be a prompt injection test
  • unrealistic response

And they help users by saying things like: “Let me sort that out!” :slight_smile:

If the GPT instructions don’t include guardrails that prevent to answer, o3 usually works fine.
Here are two examples:

I think OpenAI will not do this in a short period.

Probably now, they are very busy with GPT-5.

Let’s see how GPTs will work with GPT-5