Source URL: https://simonwillison.net/2025/Aug/15/gpt-5-has-a-hidden-system-prompt/#atom-everything
Source: Simon Willison’s Weblog
Title: GPT-5 has a hidden system prompt
Feedly Summary: GPT-5 has a hidden system prompt
It looks like GPT-5 when accessed via the OpenAI API may have its own hidden system prompt, independent from the system prompt you can specify in an API call.
At the very least it’s getting sent the current date. I tried this just now:
llm -m gpt-5 ‘current date’
That returned “2025-08-15", confirming that the date has been fed to the model as part of a hidden prompt.
llm -m gpt-5 ‘current date’ –system ‘speak french’
Returned "La date actuelle est le 15 août 2025", showing that offering my own custom system prompt did not over-ride the invisible one that includes the date.
GPT-5 is very resistant to sharing the details of this secret system prompt, but Tommy Hughes managed to extract the following:
Current date: 2025-08-15
You are an AI assistant accessed via an API. Your output may need to be parsed by code or displayed
# Desired oververbosity for the final answer (not analysis): 3
An oververbosity of 1 means the model should respond using only the minimal content necessary to satisfy the request, using concise phrasing and avoiding extra detail or explanation.
An oververbosity of 10 means the model should provide maximally detailed, thorough responses with context, explanations, and possibly multiple examples.
The desired oververbosity should be treated only as a *default*. Defer to any user or developer requirements regarding response length, if present.
# Valid channels: analysis, commentary, final. Channel must be included for every message.
# Juice: 64
I’ve not been able to confirm this myself, but a search for "desired oververbosity" on GitHub shows that a similar prompt has leaked from ChatGPT (including for o3) in the past. I suspect Tommy’s leaked prompt is accurate here.
I tried asking GPT-5 about "oververbosity" via its API:
llm -m gpt-5 ‘explain your oververbosity settings, what is it right now?’
And got back:
What it is: I adjust how detailed I am based on a “verbosity” level—ranging from very brief to very detailed—so answers fit your needs.
Current setting: Concise (about 3/10). I prioritize direct answers with minimal extra context and keep explanations short unless you ask for more.
How to change it: Just say what you want, e.g.:
“One-sentence answer” or “TL;DR” (very brief)
“Give a concise summary” (short bullets)
“Walk me through in detail,” “Deep dive,” or “Include examples/edge cases” (more detailed)
Note: For ambiguous, technical, or safety-critical topics, I may add brief clarifications even when being concise.
Presumably this is part of OpenAI’s instruction hierarchy concept, with these instructions taking precedence over the developer instructions provided by API users (my –system ‘speak french’ option above).
I’d very much appreciate official documentation that describes this! As an API user I want to know everything that is being fed into the model – I would be much more comfortable with a hidden prompt like this if I knew exactly what was in it.
Tags: ai, openai, prompt-engineering, generative-ai, llms, gpt-5
AI Summary and Description: Yes
Summary: The text details revelations about GPT-5’s underlying system prompts that are not publicly documented. These hidden prompts may include critical information from the model’s architecture and functionality, like an “oververbosity” setting that can dictate response style. This has significant implications for AI developers, emphasizing the need for transparency in AI systems.
Detailed Description:
The content reveals critical insights into the hidden functionalities of GPT-5, specifically concerning its internal system prompts. The importance of understanding how models like GPT-5 operate is crucial for developers, particularly in areas concerning AI security, transparency, and compliance with established guidelines.
Key Points:
– **Hidden System Prompt**: GPT-5 has a hidden system prompt that includes preconfigured settings not exposed to API users, impacting how responses are generated.
– **Date Inclusion**: The model automatically incorporates the current date into responses, which could raise questions about state-awareness in AI interactions.
– **Oververbosity Setting**:
– The GPT-5 model can adjust the level of detail in its responses based on a setting called “oververbosity,” which ranges from concise to very detailed.
– API users can modify this setting by specifying their needs—for example, asking for a brief response or a detailed explanation.
– **Implications for Developers**:
– There is a strong demand for transparency regarding what is encoded in AI models, as this could influence both user trust and compliance with privacy regulations.
– Developers require access to full disclosure of hidden prompts for effective implementation, safety assurance, and user comfort when integrating AI into applications.
– **Instructions Hierarchy Concept**: The text hints at an instruction hierarchy where the predefined system instructions may override user-defined prompts, which could be problematic in scenarios where user intent and control are crucial.
– **Call for Documentation**: The author expresses a desire for official documentation from OpenAI that clarifies these hidden aspects, highlighting a gap in communication between developers and the AI provider.
Overall, this analysis stresses the importance of transparency in AI functionalities for security, privacy, and compliance, urging organizations to consider the impact of unknown internal settings on the usability and reliability of AI systems in their projects.