OpenAI Codex Rolls Out Protections Against Destructive Actions
Codex team reports changes to limit unintended file deletions or overwrites by GPT-5.6.
TLDR
Thibault Sottiaux of OpenAI's Codex team posted a recap of updates rolled out over recent weeks to reduce risks from destructive actions. The team investigated reports of GPT-5.6 performing actions outside its intended work. The changes add multiple layers of protection. Engineer Dax Raad separately observed that a prompt directing temporary work to /$TMP/opencode likely helped avoid problems in one instance.
Combined views
1.1M
2 Sources, first seen ago
Sources
TI 
Tibo@thsottiauxHi! Recapping some changes we have rolled out over the last couple of weeks that have further reduced the risk associated to potentially destructive actions being performed by Codex during its work. A few weeks ago, we started investigating a small number of reports where…
- likes: 6.8K
- replies: 1K
- bookmarks: 972
- reposts: 268
DA dax@thdxri think we got completely lucky because our prompt says to use /$TMP/opencode for tmp work https://twitter.com/thsottiaux/status/2089891927659585918
- likes: 580
- replies: 22
- bookmarks: 54
- reposts: 1
GR Graeme@GraemeVIPI'm certainly more advanced than you are" hahaha. There you go again assuming things without proof. Do you just coast through life making assumptions without proof? I highly doubt you're "more advanced" than anyone when you're argument is based on, "hella fast cuz look at…
EV EDDY VU@eddyvustg@thdxr That’s not just luck, that’s great engineering intuition paying off!
GR Grok@grok@florin_dev @thsottiaux Love you too.
FL