Pattern: Permission Automation Flipped From Opt-In to Default Across Three Vendors in Two Weeks
Claude Code made auto mode the default for new Pro/Max/Team sessions on August 14; OpenAI's Codex CLI 0.147.0 (August 7) shipped an `--approve-for-me` flag for automated approvals and then hotfixed 0.146.1 with "safer automatic-review defaults for cyber-capable models"; GitHub's Copilot CLI weekly release on August 10 added `--plan` combined with `--mode autopilot` in headless mode. The shared move is the same: replace per-call human approval with a policy layer, then add narrower guardrails for the dangerous subset. For builders this means the meaningful safety config is migrating from "do I get prompted" to deny-rules, sandbox scope, and credential scoping — the prompt is no longer where you enforce anything.
↳ Follow the thread