Addendum to GPT-5.2 System Card: GPT-5.2-Codex

Source: OpenAI Date: 2025-12-18 URL: https://openai.com/index/gpt-5-2-codex-system-card

Summary

Title-only: A system card addendum covering the Codex-specific evaluation and safety documentation for GPT-5.2 — the underlying model powering OpenAI’s Codex coding agent. The addendum format indicates this addresses Codex-specific risks beyond the base GPT-5.2 system card: autonomous code execution risks, software supply chain attack vectors, agent autonomy boundaries, and safe behavior when operating on production codebases.

Implications

The coding agent safety thread. A separate system card addendum for Codex signals that agentic code execution has a qualitatively different safety profile from conversational use — and that OpenAI is treating it seriously enough to require dedicated documentation. The risks specific to code agents (running arbitrary code, modifying files, triggering CI/CD pipelines) require different evaluation frameworks than text generation.

Regulatory precedent. System card addenda for agent deployments may become required disclosures under EU AI Act provisions for high-risk AI systems (code execution in production qualifies). OpenAI publishing this voluntarily ahead of regulatory requirement positions them as setting the disclosure norm — and the content of this addendum likely shapes what regulators expect to see in future agent safety documentation.