GPT-4 System Card
webCredibility Rating
High quality. Established institution or organization with editorial oversight and accountability.
Rating inherited from publication venue: OpenAI
The GPT-4 System Card is a key industry document illustrating how a leading AI lab approaches pre-deployment safety evaluation for a frontier model; useful for understanding practical risk assessment and mitigation processes.
Metadata
Summary
OpenAI's system card for GPT-4 documents safety evaluations, risk assessments, and mitigations conducted prior to deployment. It covers findings from red-teaming exercises, evaluations of harmful content generation, cybersecurity risks, and potential for misuse, alongside the safeguards implemented. The document represents OpenAI's pre-deployment safety process for a frontier model.
Key Points
- •Documents GPT-4's potential risks including harmful content, disinformation, cybersecurity threats, and proliferation of dangerous knowledge.
- •Describes red-teaming efforts with external domain experts to identify safety gaps before public release.
- •Outlines mitigations including RLHF fine-tuning and rule-based reward models (RBRMs) to reduce harmful outputs.
- •Evaluates GPT-4 on uplift potential for chemical, biological, radiological, and nuclear (CBRN) threats.
- •Acknowledges residual risks and limitations of current safety measures, including jailbreaks and prompt injection.
Cited by 2 pages
| Page | Type | Quality |
|---|---|---|
| Alignment Research Center | Organization | 57.0 |
| Red Teaming | Research Area | 65.0 |
Cached Content Preview
OpenAI # 404 A doorway of mist Practical and strange at once Go with both your shoes by [gpt-5.4-thinking(opens in a new window)](https://chatgpt.com/?model=gpt-5.4-thinking&openaicom-did=3fad6070-4294-4587-8745-87bf499ab78e&openaicom_referred=true)
e09fc9ef04adca70 | Stable ID: ZjAxZjFmZm