Finding / codex
2026-08-10-codex-approve-for-me-puts-a-model-not-a-human-in-the-approval
--approve-for-me puts a model, not a human, in the approval seat.
A new --approve-for-me flag on the interactive and exec commands routes approval requests through automatic review instead of prompting the operator. It sets approval_policy="on-request" with the workspace-write sandbox, and propagates across root, exec, resume, and fork argument handling while preserving later subcommand permission overrides.
Channel: tagged-release. Ancestry: grep of the rust-v0.146.0...rust-v0.147.0 compare commit list finds exactly one match: b7a61066081644e0d8b2c0b4dbfd7408ac1514df 'Add an --approve-for-me CLI flag (#36373)'. PR #36373 merged 2026-07-31T18:28:24Z; first non-prerelease tag containing it is rust-v0.147.0.
Operator consequence: Watch, and decide deliberately before enabling it in CI. This is the flag that converts Codex's approval prompt from a human checkpoint into a model checkpoint on a single command-line switch, and it also silently pins the sandbox to workspace-write. If your governance story rests on 'a person approved every escalation', add --approve-for-me to whatever you lint agent invocations with. Note the ordering trap: subcommand permission overrides still win, so the effective policy is not always what the flag implies.
Receipt
Finding metadata
Run: 2026-08-10-weekly-digest-2026-08-03_2026-08-10-frontier-v0
Finding ID: 2026-08-10-codex-approve-for-me-puts-a-model-not-a-human-in-the-approval
Accepted signals
Source links
Primary links, including exact changelog lines when available.