OpenAI has released an addendum to the GPT-5.2 System Card, specifically detailing safety protocols for GPT-5.2-Codex. This new documentation highlights a dual-layered security approach designed to mitigate risks associated with advanced AI coding agents. The release outlines critical model-level mitigations, including specialized safety training intended to prevent the generation of harmful code or responses to prompt injections.

Furthermore, it details robust product-level safeguards, such as agent sandboxing and configurable network access, which are essential for isolating potentially risky operations. These measures are crucial for developers and enterprises leveraging Codex for software development, as they address the growing concerns around AI autonomy and security vulnerabilities. By implementing these rigorous controls, OpenAI aims to balance the high utility of autonomous coding agents with the necessary safety guardrails to prevent misuse and ensure secure deployment within professional environments.