This policy applies to the bproject07/Codex-Source fork and its fork-specific changes. This repository is an independent derivative of openai/codex; it is not an official OpenAI support or security-reporting channel.

Security fixes are applied to the current default branch. Older commits, development branches, locally modified builds, and third-party binaries may not receive fixes.

Do not open a public issue, discussion, or pull request for a suspected vulnerability.

Use GitHub's private vulnerability reporting for this repository:

Report a vulnerability privately

Include, when possible:

• the affected commit, build, and operating system;
• the affected component and configuration;
• reproduction steps or a minimal proof of concept;
• the expected impact and any known mitigations; and
• whether the issue has been disclosed elsewhere.

Do not include access tokens, account credentials, private prompts, or other people's data. The maintainers will coordinate remediation and disclosure through the private advisory.

If the issue affects an OpenAI service, account, or unmodified upstream Codex rather than this fork, use OpenAI's security policy instead.

Codex Raw's HTTP API defaults to token-free loopback operation for trusted local clients. It supports a shared bearer token and refuses non-loopback listeners without one. The token protects /readyz and /v1; /healthz intentionally remains open. This is not per-user authorization, and the built-in server uses plain HTTP. For network use, place it behind a hardened reverse proxy that enforces TLS, connection/header timeouts, and connection limits; restrict the Raw port with a firewall and prevent direct public access. See the Codex Raw security essentials for the complete operating guidance.

Users remain responsible for reviewing commands, protecting local state, and using sandbox, approval, and network controls appropriate to their environment.