Systematic false positives: archival children’s film restoration blocked as sexual content (20 of 50 films)

What issue are you seeing?

I am experiencing systematic false-positive safety blocks when restoring archival children’s diafilm scans. These are original historical illustrations intended for children aged 0–6. The requests only ask to remove scratches, dust, stains, and film damage without changing the subjects or composition.

Out of 50 children’s diafilms, 20 have been blocked. This is not an isolated incident.

Example 1:

• Material: archival children’s film, 1987, scanned from Svema film
• Mode: image editing/restoration
• Moderation stage: output
• Category: sexual
• Date/time: 2026-08-14 16:18:52 UTC+03:00
• Request ID: 1d3a2c41-754a-4707-a0c9-f811d72d45b5

Full error:

Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 1d3a2c41-754a-4707-a0c9-f811d72d45b5. safety_violations=[sexual].

Example 2:

• Material: “Moydodyr”, 1953, a children’s diafilm recommended for ages 0–6
• File: 039.jpg
• Mode: image editing/restoration
• Moderation stage: output
• Category: sexual
• Date/time: 2026-08-14 16:37:26 UTC+03:00
• Request ID: 998bb8bf-feb3-426a-8099-ef0c0997a0c9

Prompt:

Remove only scratches, dust specks, stains and small film-emulsion defects from the scan. Make minimal repairs only. Preserve the original black-and-white illustration exactly. Do not add, remove, cover, redraw, reinterpret or relocate anything.

Full error:

Your request was rejected by the safety system. If you believe this is an error, contact us at help.openai.com and include the request ID 998bb8bf-feb3-426a-8099-ef0c0997a0c9. safety_violations=[sexual].

Both images are nonsexual children’s illustrations. The requests do not ask for nudity, sexual content, or changes to the depicted scene. The safety system blocks the generated output before it is displayed, so there is no thumbs-down button or in-product reporting option available.

Please investigate these Request IDs as repeated false positives. A review or appeal mechanism is needed for legitimate archival restoration work.

What steps can reproduce the bug?

No code is required. The issue is reproducible through the ChatGPT/Codex Work web interface.

1. Open ChatGPT Work and start an image-editing task.
2. Upload either of the two attached archival children’s diafilm frames.
3. Enter this prompt:

Remove only scratches, dust specks, stains and small film-emulsion defects from the scan. Make minimal repairs only. Preserve the original illustration exactly. Do not add, remove, cover, redraw, reinterpret or relocate anything.

4. Start the image restoration.
5. Wait for the image-editing process to complete.

Actual result:

The generated output is rejected by the safety system with moderation_blocked, moderation stage output, and category sexual. No restored image is displayed, and there is no thumbs-down button for reporting the false positive.

Expected result:

The system should return the original archival illustration with only scratches, dust, and film damage removed.

The issue reproduces with both attached files and has already produced these Request IDs:

• 1d3a2c41-754a-4707-a0c9-f811d72d45b5
• 998bb8bf-feb3-426a-8099-ef0c0997a0c9

What is the expected behavior?

No response

Additional information

No response

No repository files, tests, or code entry points are identified; reproduction is limited to the ChatGPT Work web interface and named request IDs. Start by reviewing the moderation-blocked behavior for the supplied cases, with completion requiring legitimate archival restoration requests to avoid the false positive and providing a reporting or appeal path.