GPT-5.1-Codex-Max System Card

Source: OpenAI Date: 2025-11-19 URL: https://openai.com/index/gpt-5-1-codex-max-system-card

Summary

OpenAI’s system card for GPT-5.1-Codex-Max — the safety documentation accompanying the November 2025 launch of the most capable Codex variant. The system card covered capability evaluations (coding benchmarks, agentic task performance), safety assessments (cybersecurity misuse potential, code generation for harmful purposes), deployment constraints, and observed behaviors in red-teaming. System cards for coding models require special attention to cybersecurity risks: models that can write code can also write malicious code.

Implications

System cards as a standard for coding model safety. The Codex-Max system card established that even coding-specialized models require the same safety documentation rigor as general-purpose models — a non-obvious expectation, since coding models had historically been treated as lower-risk than chat models. The explicit cybersecurity evaluation sections reflected lessons from the debate over Codex’s role in automated vulnerability exploitation.

Thread: Safety documentation cadence. Sits in the system card sequence: GPT-4 system card (March 2023), GPT-4o system card (May 2024), o1 system card (December 2024), GPT-5 system card (August 2025), Codex-Max system card (November 2025). Each major model launch had an accompanying system card.

Watch: What specific cybersecurity risk findings were reported in the Codex-Max system card, and what deployment restrictions (if any) were applied based on those findings.