June 28, 2026
Secret files, public meltdown
A way to exclude sensitive files issue still open for OpenAI Codex
A year later, users are still begging for a simple ‘don’t read my secrets’ button
TLDR: A fresh request asks OpenAI Codex for a built-in way to avoid reading private files, but the feature still isn’t there. Commenters are torn between outrage that it’s unresolved after a year and mockery that any software promise here would be a dangerous illusion.
The latest OpenAI Codex dust-up is less about shiny new features and more about a very human panic: please stop the bot from snooping on private files. One user revived an old request asking for a simple, shared way to mark certain files as off-limits — think passwords, secret keys, and personal login folders — so the coding assistant won’t read them or send them away by accident. Sensible? The crowd is split between “how is this still not fixed?” and “this is fake safety and everyone knows it.”
The most exasperated reaction came fast: “it has been a year and still it is not resolved,” basically the software version of staring into the camera on a reality show. Another commenter wanted this turned into a universal rulebook across tools, comparing it to shared community standards and asking what rival systems do. In other words: if everyone agrees this is scary, why is there still no normal, obvious solution?
But then the thread turned spicy. Critics called the idea “snake oil” and “pointless,” arguing that if your app can reach the file, the assistant may still stumble into it somehow. Their advice was brutally old-school: lock files down with system permissions, or isolate the tool in a container so the files simply aren’t there. The unspoken meme of the thread: users want a magic “don’t peek” sticker, while skeptics are yelling, “put it in a locked drawer!” For now, the feature request is open — and the comments are serving pure trust-issues theater.
Key Points
- •The article requests a mechanism for OpenAI Codex to explicitly exclude sensitive files and paths from being read or sent to the model.
- •The proposed design includes both repository-level and global ignore configurations, such as a local `.codexignore` plus a global ignore file.
- •Example sensitive targets listed for exclusion include `.env`, `.env.*`, `.pem`, `id_`, `.aws/`, and `.ssh/`, while some directories like `node_modules/` may remain searchable.
- •The submitter says they are willing to implement the feature and add tests.
- •A related issue, #205, was closed in favor of codex-rs, but the article states that as of 2025-08-28 a comparable feature does not appear to exist in codex-rs.