[ChatGPT for Chrome] macOS repeatedly blocks Pillow _imaging module during PDF generation

Open 💬 1 comment Opened Jul 20, 2026 by eyedroot

What issue are you seeing?

While using ChatGPT for Chrome with Full access to generate a PDF from the currently open webpage, the ChatGPT/Codex task created a Python 3.12 virtual environment and installed Pillow 12.3.0 via uv.

During the task, macOS repeatedly displayed this security warning:

“_imaging.cpython-312-darwin.so” Not Opened Apple could not verify “_imaging.cpython-312-darwin.so” is free of malware that may harm your Mac or compromise your privacy.

The affected module was located at:

~/Documents/Codex/2026-07-20/new-chat/work/.venv312/lib/python3.12/site-packages/PIL/_imaging.cpython-312-darwin.so

The macOS unified log showed:

AMFI: '.../PIL/_imaging.cpython-312-darwin.so' is adhoc signed.
The file is adhoc signed or signed by an unknown certificate chain.

The warning appeared repeatedly and was alarming enough that I clicked Move to Trash, which removed the native module. The PDF was ultimately generated, but the temporary Python environment was left broken.

Before deletion, the module's SHA-256 matched Pillow's installed RECORD entry, so this appears to be a signing/provenance handling problem rather than a package-integrity mismatch.

What steps can reproduce the bug?

  1. On macOS, open an article in Google Chrome.
  2. Open the ChatGPT for Chrome sidebar.
  3. Select 5.6 Sol High with Full access.
  4. Ask ChatGPT to translate/summarize the current article and generate a PDF while preserving code examples.
  5. Wait while the task creates a local Python environment and renders the PDF.
  6. Observe repeated macOS Gatekeeper/AMFI warnings for Pillow's _imaging.cpython-312-darwin.so.

The task took approximately 10 minutes before returning the generated PDF.

What is the expected behavior?

Native dependencies installed or executed by ChatGPT/Codex should not trigger repeated macOS Gatekeeper warnings. The runtime should use a packaging or execution path that macOS accepts, or fail gracefully with a clear explanation instead of presenting the user with a frightening system security dialog.

Additional information

  • ChatGPT for Chrome: v1.2.27221.15725
  • Google Chrome: 150.0.7871.125
  • macOS: 26.5.2 (25F84)
  • Platform: Darwin 25.5.0 arm64 arm
  • Python: 3.12
  • Pillow: 12.3.0
  • Installer recorded by package metadata: uv
  • Removed module SHA-256: ee754300d3bda6413b80ddabe3a7ed7b419e1b0468bf2830e2b19fe68f14be42

A possibly related issue reports the same class of macOS warning for a different native binary: https://github.com/openai/codex/issues/28190

I am not attaching the original screenshot because it includes unrelated work tabs and account information. The exact warning text and relevant diagnostics are included above.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗