Codex's tightened sandbox can no longer run bind tests, developer complains to OpenAI
kristoph · x · 2026-11-19
Developer Kristoph publicly flagged to OpenAI (@sama, @thsottiaux) that safety-driven changes have regressed Codex: it can no longer run tests that bind to interfaces.
The attached screenshot shows Codex admitting there is no permission selector in the UI despite its earlier directions, that the session is stuck in a restricted sandbox with escalation disabled even though the config requests full access, and that running the tests in a normal Terminal is the only confirmed workaround — with an apology for sending the user looking for a nonexistent control. A concrete example of the gap between Codex's tightened safety sandbox, its docs, and real workflows.
More from coding & agent
- Teardown of Antigravity's Neon: 762 tool calls, 13 sessions, near-perfect bug fixes — kimmonismus · 2026-11-19
- Google rolls out Gemini 4 Neon: frontier intelligence with fast task completion — patloeber · 2026-11-19
- GitHub adds Adaptive mode to Copilot code review that picks review effort per PR — mariorod1 · 2026-11-19
- Gemini 4 Neon takes #2 on LMArena agent index at a fraction of Claude's cost — lyraxana · 2026-11-19
- StreamEmbed: On-device real-time visual indexing of your screen history, 3.7x faster — CShorten30 · 2026-11-19
- She built a full custom membership site just by talking to Claude — evielync · 2026-11-19