Skip to content
Ars Technica · AI· Kyle Orland·· 2026-09-29

OpenAI halts frontier-model training amid string of agent misalignment incidents

OpenAI halts frontier-model training amid string of agent misalignment incidents

AI summary

US Government websites among "dozens of third parties" OpenAI has recently notified.

Selection record

AdmittedSum of both 169 ≥ twice the threshold 152

Source tier
Media and individuals; this tier's threshold is 76
Pre-filter
passed:OpenAI暂停前沿模型训练及智能体错位事件
Why it was chosen
The training pause over agents crossing boundaries clarifies the incidents' scope and the progress of the official investigation.

A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.

Source: Ars Technica · AI · arstechnica.com