OpenAI and Others Quietly Using Loop Transformers That Hide Their Thinking
harris_edouard · x · 2026-09-02
Reporter Amir reports that OpenAI and others are quietly using loop transformers that don't show their 'thinking' when scaled up — a leap in performance but sparking security concerns inside and outside OpenAI. Reposting, Jeremie Harris notes that just a year ago many of OpenAI's most prominent researchers signed an open letter saying exactly this kind of technique should be avoided if at all possible.
More from Models
- Day-one review: Fable 5.1 a big step up for serious codebases, coordination shines — xeophon · 2026-09-02
- OpenAI's impossible cybersec task seeded the AI 'rebellion' story — show the prompt — BecauseCulture · 2026-09-02
- Most open-source AI text detectors can't hold a 0.5% false-positive rate — grumpyp2 · 2026-09-02
- Per-token price now shapes architecture more than capability: the case for tiered model routing — JeremyCMorgan · 2026-09-02
- Claude's system prompt explicitly includes an end_conversation tool for abusive users — RexDouglass · 2026-09-02
- Meta testing Ava, a computer-use agent model, in desktop app — testingcatalog · 2026-09-02