A way to exclude sensitive files issue still open for OpenAI Codex

What actually happened with OpenAI Codex?

A developer opened a request on OpenAI's Codex GitHub repository on 28 August 2025, asking for a simple ".codexignore" file, similar to the .gitignore file most developers already use. It would stop the Codex AI coding assistant from reading sensitive files such as .env files, SSH keys, .pem certificates and AWS credential folders.

Nearly a year later, that issue is still open, still unassigned, and OpenAI has not officially responded. A near identical request was raised back in 2025 under issue #205, closed when OpenAI rewrote Codex in Rust, on the promise the new version would handle it. It did not.

Does OpenAI just ignore security issues?

No, and that is what makes this case worth watching. When Check Point reported a separate critical Codex CLI flaw (CVE-2025-61260) to OpenAI on 7 August 2025, OpenAI shipped a fix in just 13 days, in Codex CLI version 0.23.0. So the company can move fast when a flaw is formally classified as a vulnerability.

The ".codexignore" request has not been given that classification. It sits in a grey area: not a bug, more a missing safety feature, and those tend to queue behind flashier priorities. The Hacker News discussion of the issue drew around 120 points and 83 comments, several pointing out a harder truth: even a proper ignore file would not fully solve this, because an AI agent can still leak a secret indirectly, by running a shell command like cat or rg that prints the contents of a file it was never meant to open directly.

Why should a UK small business owner care about a GitHub issue?

Because you are increasingly the customer of the software this affects, even if you never write a line of code yourself. If you have paid a developer, a freelancer, or an agency to build your booking system, website backend or automation using AI coding tools like Codex, your live API keys, payment credentials and customer database logins may have passed through exactly the kind of file this issue is about.

This is not a hypothetical risk. GitGuardian's 2026 State of Secrets Sprawl report found 28.65 million new hardcoded secrets exposed in public GitHub commits during 2025, a 34% year on year rise. Commits made with AI coding assistance leaked secrets roughly 3.2% of the time, more than double the 1.5% rate for commits written by hand.

Could this actually affect a business as small as mine?

Yes, and the smaller you are, the more it matters, because you likely have no in house developer checking what gets committed. A leaked Stripe key, Google API key or email password sitting in a public repository can be found and abused within hours, not months.

What can I actually do about this right now?

You do not need to understand Rust or GitHub internals to protect yourself. Ask whoever builds or maintains your systems these three things directly:

• Confirm that .env files, credential files and API keys are never committed to any code repository, checked with a tool like git-secrets or GitGuardian before every commit.
• If they use Codex or a similar AI coding agent, ask them to run it inside a restricted sandbox profile (in ~/.codex/config.toml) or a container that only mounts the specific project folders it needs, not your whole system.
• Rotate any API keys or passwords that have ever touched a shared or AI assisted codebase, and set a reminder to do this again every few months.

None of this requires you to become technical. It requires you to ask direct questions and expect a direct answer, the same way you would ask a builder whether the wiring meets regulations.

What is Braynex Services' view on this?

Our position is straightforward: this issue is a symptom of a bigger pattern, not a one off. Across the businesses we work with, from a Leeds salon to a Manchester restaurant, the biggest financial risks are usually invisible ones: missed calls, platform commissions eating into margins, and now, credentials quietly exposed by tools built for speed rather than caution.

Owning your own website, booking system and data, rather than renting space on someone else's platform, already puts you ahead. But ownership only pays off if what you own is built securely. An AI coding assistant that cannot yet reliably tell the difference between a normal file and your SSH key is a reminder that "built with AI" is not automatically "built safely."

If you are not sure what is sitting exposed in your own systems, or whether your booking platform, website or automation setup was built with these risks in mind, book a free audit with Braynex Services at braynexservices.com. We will tell you plainly what we find, no jargon, no upsell.