Justin McKelvey
Ran snowboard shops. Taught myself to code. Now I build AI that does real work.
Codex Security (2026): What OpenAI's Code Scanner Does, Which Plan Runs It, and How to Use It on an AI-Built App
Quick Answer
Codex Security is OpenAI's vulnerability scanner for your code. It runs four ways: a plugin inside Codex for local scans, a CLI for repeat scans and CI, an SDK, and Daybreak Defense Factory for connected GitHub repositories. The GitHub version is listed for Enterprise only. OpenAI doesn't publish which plans get the local plugin or what a scan costs, and the CLI's cost limit is an estimate, not a cap. A scan never changes your code; fixes are a separate step you approve.
From OpenAI's Codex Security docs and Codex pricing page, read November 21-22, 2026 · Author: Justin McKelvey, AI consultant, former CTO at an AI startup, 50+ products shipped
What Codex Security is
Every vibe-coding tool now says "secure." Codex Security is OpenAI admitting the code still needs checking. It reads an application's code, flags likely vulnerabilities with the evidence for each one, and can propose a patch and test whether the patch actually closes the hole. It scans your app's code; it's not about what Codex itself is allowed to do on your machine, which is a separate set of settings.
Google searches for it went from about 170 in February 2026 to 3,600 in March and still run around 1,300 a month, a lot of them from people who built something with AI and now want to know if it's safe to put customers on it.
The four ways to run it, and which plan gets each
| Way in | What it's for | Plan, per OpenAI's pages |
|---|---|---|
| Codex Security plugin | Scan a local folder from the ChatGPT desktop app or Codex CLI and work through findings | Not published. Needs Codex (Plus, $20 a month, and up); managed workspaces can block it |
| Codex Security CLI | Repeatable scans from a terminal, bulk scans, checks in CI | ChatGPT sign-in or an OpenAI API key; no plan list published |
| SDK | Build scanning into your own tool | Developer use |
| Daybreak Defense Factory (Preview) | Scan connected GitHub repos, pull in other scanners' findings, propose fixes as a team | Enterprise only; not Plus, Pro or Business |
So a small business on Plus or Business is looking at the plugin, not the GitHub dashboard. What Codex itself costs on each plan, and when Pro ($100, $200 or $500 a month) starts to make sense, is on Codex pricing.
What a scan costs
OpenAI doesn't publish a price per scan. On a ChatGPT sign-in, a scan uses your Codex allowance like any other long task. On an API key, it bills at API rates. The CLI lets you set an estimated cost limit in dollars, and OpenAI's own FAQ says it's an estimate rather than a hard cap. A deep scan, which repeats the review over the same code, can run for up to 96 hours. Start with a standard scan of one folder.
Worked example: a first scan on an app you didn't write by hand
The owner: a two-person services business with a client portal built in Lovable on Supabase, synced to GitHub. Nobody on the team reads code. Here's the 30-minute routine I'd give them.
- Get the code local. Clone the GitHub repo to a laptop. (If it was never synced, export it first.)
- Install the plugin. ChatGPT desktop app, select Codex, open Plugins, search "Codex Security," enable it, open Security in the sidebar. If it says "Disabled by admin," that's your workspace owner's setting.
- Run a standard scan, not a deep one. Security > Scans > + Scan, pick the repo, leave Deep scan off.
- Add one line of context. In Additional context: "Customers log in and see only their own invoices. Check that one customer can never see another's data." That's the bug that actually hurts small apps.
- Read the report by severity. Anything high goes to whoever maintains the app. Don't apply fixes you can't explain to them.
What it won't see: settings that live outside the code. Supabase row-level security you set in the dashboard and never exported, API keys pasted into a hosting platform, a storage bucket set to public. Those are why Is Supabase secure? walks through the dashboard checks by hand.
Codex Security vs other scanners
Codex Security reasons about what the code is supposed to do, which is where AI-built apps break: the login works, but customer A can load customer B's page. Traditional scanners are better at the boring, reliable stuff, like known-vulnerable packages and leaked keys in commit history, and many are free. I'd run both. The side-by-side of the scanners worth using on an AI-built app is in best security scanners for vibe-coded apps.
My take
I haven't put Codex Security through a client codebase yet, so this is how I'd use it, not a review.
A scanner is a smoke detector. It's worth having and it's not an inspection. The apps I see in rescue work don't fail on exotic attacks; they fail on access rules nobody wrote down, and a scanner only finds those if you tell it what "only their own data" means. That one line of context in step 4 is the whole trick.
The plan gap bugs me. OpenAI puts the team dashboard behind Enterprise and doesn't say who gets the local plugin, which means the small businesses most likely to ship an unreviewed AI-built app are the least sure they can run it. Try the plugin on whatever plan you have; if it installs, scan this week. The setup walk-through for Codex itself is in how to use Codex.
What to watch out for
- The cost limit isn't a cap. Watch the first scan on an API key instead of walking away.
- Deep scans run long. Up to 96 hours. Not your first move.
- A clean report isn't a clean app. Dashboard settings and hosting keys are outside what it reads.
- Fixes are code changes. Proposed patches still need someone who can tell you what they changed and test the app after.
If the scan comes back red and nobody on your team can read the fixes, that's the situation my vibe code rescue work is for.
Frequently asked questions about Codex Security
What is Codex Security?
Codex Security is OpenAI's code vulnerability scanner. It reads your application's code, reports findings with evidence and suggested fixes, and can generate a patch and check whether it works. You run it as a plugin inside Codex, from a terminal with the Codex Security CLI, through an SDK, or in Daybreak Defense Factory for connected GitHub repositories.
Is Codex Security free?
Not as a separate free product. The local plugin and CLI run on a ChatGPT sign-in or an OpenAI API key, so scans draw on your Codex usage or your API bill. OpenAI doesn't publish a per-scan price; the CLI has an estimated cost limit you can set, and OpenAI's own FAQ says that limit is an estimate, not a hard cap.
Which ChatGPT plans include Codex Security?
Scanning connected GitHub repositories through Daybreak Defense Factory is listed on OpenAI's Codex pricing page for Enterprise only, not Plus, Pro or Business, as of November 22, 2026. OpenAI doesn't publish a plan list for the local plugin or CLI; Codex itself starts on ChatGPT Plus at $20 a month. In a managed workspace, an admin can disable the plugin.
Does Codex Security change my code?
A scan doesn't. OpenAI says the scan leaves your source unchanged and saves a report with evidence and remediation guidance. Fixes are a separate step where Codex proposes a patch and checks whether it resolves the finding, and you decide whether to apply it.
Can Codex Security scan a Lovable, Bolt or Replit app?
Yes, if you can get the code into a local folder or a GitHub repository. Export or sync the project to GitHub, clone it, and run the plugin on the folder. It checks application code, so it won't see settings that live only in a hosting dashboard, such as Supabase row-level security policies you never exported or keys pasted into a platform's settings.
Sources
- OpenAI, Codex Security overview, read November 22, 2026: the four ways to run it; scans application code.
- OpenAI, Codex Security plugin quickstart, read November 22, 2026: install steps, scans leave source unchanged, Daybreak as a separate plugin, admin-disabled plugins.
- OpenAI, Codex Security CLI quickstart and FAQ, read November 21-22, 2026: ChatGPT or API-key sign-in, estimated cost limit, deep scan run time.
- OpenAI, Codex pricing, read November 21-22, 2026: "Codex Security for connected GitHub repositories" by plan.
Ship AI-built code without the 2am surprise
The free 20-point security checklist I run on every AI-built codebase before it touches real users.
Frequently Asked Questions
- What is Codex Security?
- Codex Security is OpenAI's code vulnerability scanner. It reads your application's code, reports findings with evidence and suggested fixes, and can generate a patch and check whether it works. You run it as a plugin inside Codex, from a terminal with the Codex Security CLI, through an SDK, or in Daybreak Defense Factory for connected GitHub repositories.
- Is Codex Security free?
- Not as a separate free product. The local plugin and CLI run on a ChatGPT sign-in or an OpenAI API key, so scans draw on your Codex usage or your API bill. OpenAI doesn't publish a per-scan price; the CLI has an estimated cost limit you can set, and OpenAI's own FAQ says that limit is an estimate, not a hard cap.
- Which ChatGPT plans include Codex Security?
- Scanning connected GitHub repositories through Daybreak Defense Factory is listed on OpenAI's Codex pricing page for Enterprise only, not Plus, Pro or Business, as of November 22, 2026. OpenAI doesn't publish a plan list for the local plugin or CLI; Codex itself starts on ChatGPT Plus at $20 a month. In a managed workspace, an admin can disable the plugin.
- Does Codex Security change my code?
- A scan doesn't. OpenAI says the scan leaves your source unchanged and saves a report with evidence and remediation guidance. Fixes are a separate step where Codex proposes a patch and checks whether it resolves the finding, and you decide whether to apply it.
- Can Codex Security scan a Lovable, Bolt or Replit app?
- Yes, if you can get the code into a local folder or a GitHub repository. Export or sync the project to GitHub, clone it, and run the plugin on the folder. It checks application code, so it won't see settings that live only in a hosting dashboard, such as Supabase row-level security policies you never exported or keys pasted into a platform's settings.
More on Vibe Code Rescue
Neon Pricing (2026): Free Plan Limits, $0.106 per CU-Hour, and When Scale-to-Zero Stops Saving You Money
Neon's Free plan gives each project 100 compute hours a month, and its paid Launch plan bills $0.106 per CU-hour plus $0.35 per GB of storage. A small database that sleeps when idle costs about $15 a month; the same database left on all month is about $21, and on the Free plan it gets suspended around day 17.
Supabase Pricing (2026): Free, Pro $25, Team $599, and the Line Item That Jumps Your Bill
Supabase is free for 2 small projects, $25 a month on Pro and $599 on Team. Most Pro apps pay exactly $25, until monthly active users pass 100,000: then each extra user costs $0.00325, and a $25 bill becomes $204.
DigitalOcean Pricing (2026): Droplets from $4, App Platform from $5, and the Same App Priced Three Ways
DigitalOcean Droplets start at $4 a month and App Platform containers at $5, but the real question is which way to run your app. The same small app with a database costs $14.40, $29.55 or $39.15 a month depending on how much DigitalOcean manages for you.
Sentry Pricing (2026): Free, Team $26, Business $80, and What Overages and Seer Add
Sentry is free for one developer, then $26 a month (Team) or $80 (Business) billed annually, plus pay-as-you-go overages. A worked bill for a 3-developer app: $67 on Team, $177 on Business, and $187 on Team once the Seer AI debugger is added.
Written by
Justin McKelvey
I spent 12 years running snowboard and mountain-bike shops in Colorado, then taught myself to code. Since then: 50+ products shipped, Head of Product at a national sports platform, CTO at an AI startup. Now I put AI to work in real businesses in Austin, TX and remote, and write down what actually works.
Work with meWant more help? Get the free security checklist · Get a free 20-min repo audit