OpenAI makes Codex Auto‑review free for every ChatGPT sign‑in
A second agent now watches Codex's first one for risky actions, free for anyone signed in with ChatGPT. The Codex poster @thsottiaux told users to run it instead of full access.
OpenAI has made Auto-review free for all users who sign in to Codex through a ChatGPT account, the Codex-focused X account @thsottiaux said on X on Tuesday. He labelled the post "Day 2.1", part of a numbered series. TestingCatalog, an outlet that tracks AI product changes, reported the change within minutes. OpenAI has not published a separate announcement that we found.
@thsottiaux said users switch it on under Settings, Permissions, Auto-review. He described it as an improvement on the default sandbox, which asks the user to approve everything and is "prone to decision fatigue" unless the user spends a lot of time configuring specific rules. These descriptions are the company's own, and nobody outside OpenAI has yet tested how well the feature works.
https://x.com/thsottiaux/status/2107368734981517634
The feature lets people run long tasks while a second agent reviews every action the primary agent takes, according to @thsottiaux. He said its only goal is to stop high-risk actions and to guard against unwanted actions that do not match the original user intent. Which actions the reviewer counts as high-risk was not detailed in the post.
In a follow-up post @thsottiaux told users to "use auto-review instead of full-access", adding that it is "not a trade-off anymore". TestingCatalog was more cautious, noting that approving small steps without granting full access has been a challenge and that it will have to prove it makes the process less distracting.