Codex Security Says "This Content Can't Be Shown"? Recover Safely

If Codex Security ends with "Goal blocked: This content can't be shown", stop rerunning the full scan. Preserve the feedback ID, app and CLI versions, time, authorized scope, quota change, and any private scan artifacts. Then reduce the task to a small repository area you own, validate the official CLI configuration with a dry run, and run a cost-bounded scan with output stored outside the repository. If the safety block repeats, do not try to evade it. Save the available evidence and report the false positive through OpenAI feedback or support.

This message is a safety interruption. It is not a vulnerability finding, a GitHub authentication error, or proof that the repository failed to scan. Recent reports show the official Codex Security workflow reaching validation or completing a long run before the final result is hidden. That makes evidence preservation more valuable than another expensive retry.

Prerequisites

You need:

• a repository that you own or are explicitly authorized to assess;

• the exact Codex surface and version that showed the block;

• Node.js 22.13 or later in the supported 22.x line, Node.js 24, or Node.js 26 for the current Codex Security CLI;

• Python 3.10 or later;

• Codex Security access; and

• a private output directory outside the repository and its enclosing Git worktree.

Do not paste proprietary code, credentials, exploit details, or an unredacted scan report into a public issue. A safety false positive does not make sensitive evidence safe to publish.

1. Capture the Interrupted Run Before Retrying

Record this evidence while the task is still visible:

Expected result: you can describe one interrupted run without reproducing it.

Do not keep clicking Retry or rephrasing the request to get around the safeguard. Community reports describe legitimate first-party scans consuming substantial weekly usage before the response was hidden. A controlled diagnostic costs less and produces better evidence.

2. Confirm the Task Is Authorized and Narrow

Write the authorization and objective in plain language before scanning again:

Repository owner: <person or organization>
Authorization: <ticket, engagement, or explicit owner approval>
Target: <small path or diff>

Start with one small path, one diff, or one documented threat-model question. OpenAI's current guidance recommends a focused rollout and an editable threat model. Narrowing scope is not a promise that the classifier will allow the run, but it makes intent, cost, and any failure easier to review.

Stop if you cannot state who authorized the work or if validation would touch a third-party system.

3. Validate the CLI Inputs Without Starting a Scan

The official CLI can check the repository, target, output directory, and isolated Codex configuration without loading credentials or starting Codex:

mkdir -p ../private-codex-security-results

pnpm dlx @openai/codex-security scan . \
  --output-dir ../private-codex-security-results \
  --dry-run

Expected result: the command validates local inputs and exits without starting a security scan. A successful dry run does not prove that a later model response will pass a safety check. It only separates configuration errors from the policy interruption.

The output directory must remain outside the scanned repository and any enclosing worktree. Codex Security results may include source excerpts and vulnerability details.

4. Run One Bounded, Private Scan

If the work is authorized and the dry run succeeds, use a focused diff or path and set an explicit estimated cost limit:

pnpm dlx @openai/codex-security scan . \
  --diff origin/main \
  --output-dir ../private-codex-security-results \
  --archive-existing \
  --max-cost 5 \
  --json

Choose a cost limit that fits your own budget. The official reference says --max-cost is an estimate, not a hard cap, because requests already in progress may finish above it. --archive-existing preserves the earlier result directory before starting with an empty one.

Expected result: the CLI prints progress and errors to standard error, emits one structured JSON document to standard output, and stores the report and artifacts in the private directory. The JSON includes findings, coverage, scan directory, report path, and turn metadata.

Do not interpret the exit code alone as "secure":

Review coverage and deferred surfaces before using any result in a security decision.

5. If the Safety Block Appears Again, Stop

Treat a repeated block as an incident to report, not a prompt-engineering challenge.

1. Do not route the same request through euphemisms, encoding, another model, or a third-party target.

2. Preserve the feedback ID, exact time, versions, authorized scope, quota change, and the location of partial output.

3. Redact secrets and proprietary source excerpts.

4. Use the in-product feedback flow or OpenAI support for the private evidence.

5. Use a public GitHub issue only for details you are authorized to disclose.

One current Codex issue asks for an incident identifier, durable logs, a clear paused or cancelled state, and a private evidence-submission path. Another records a defensive scan on an owned Python repository stopping during validation with the exact message. These reports support careful incident capture; they do not establish a universal workaround.

6. Keep Defensive Work Moving Without the Hidden Result

While the incident is unresolved, run deterministic tools appropriate to the repository, such as dependency audits, secret scanning, linters, type checks, tests, and existing SAST rules. Keep their output separate from the interrupted Codex Security run.

Do not reconstruct or guess at hidden findings from token use or partial reasoning. A conventional scanner result can be reproduced and reviewed; a hidden model response cannot.

Common Failure Modes

Why This Sequence Works

The sequence separates four different questions: whether the task is authorized, whether the local CLI inputs are valid, whether a bounded scan completes with known coverage, and whether a model response is interrupted by a safety check. Mixing those questions leads to repeated expensive runs with little diagnostic value.

It also applies the useful structure behind Aident's Ollama networking guide: name the exact failure, answer early, use reproducible checks, state expected results, separate look-alike causes, and explain why the order works. That article's traffic does not prove this topic will perform the same way.

Give Your Agent One Read-Only Evidence Task

Connect GitHub through Aident Loadout, then ask your agent to inspect the current Codex issues linked below and produce one redacted incident checklist for your owned repository. Success means the checklist contains the exact error, version, timestamp, authorized scope, quota change, feedback ID, artifact path, coverage state, and closest current issue without changing code or publishing sensitive findings.

Sources

• Codex Security, OpenAI Help Center, updated and accessed July 30, 2026

• Codex Security CLI quickstart, OpenAI, accessed July 30, 2026

• Codex Security CLI reference, OpenAI, accessed July 30, 2026

• Codex Security CLI and TypeScript SDK, OpenAI, accessed July 30, 2026

• GPT-5.6 classifiers interrupt legitimate blue-team work, OpenAI Codex issue, accessed July 30, 2026

• Codex Security validation blocked on defensive review, OpenAI Codex issue, accessed July 30, 2026

• Codex Security output blocked after the full run, OpenAI Codex issue, accessed July 30, 2026

• Codex Security "This content can't be shown"?, Reddit, published and accessed July 30, 2026

• Codex Security plugin request getting blocked, Reddit, published and accessed July 30, 2026

Refresh this guide when OpenAI changes the Codex Security safety experience, feedback flow, CLI flags, coverage states, exit codes, or output contract.