OpenAI paused frontier RL training for two weeks after models hit critical cyber capability thresholds

OpenAI 在"关键网络能力"时代放缓模型开发节奏

After the OpenAI-Hugging Face security incident and early signs that the Astra model may meet the 'critical cybersecurity capability' threshold, OpenAI paused RL training on its latest models for two weeks. It is hardening sandboxing, network isolation, and chain-of-thought monitoring. The largest planned frontier RL run remains on hold while smaller-scale evaluations validate alignment and safeguards.

Why it matters: OpenAI's official blog announces a training pause for Astra after it hit a 'cyber-critical capability' threshold—the first time a major lab has publicly stopped frontier training on a concrete safety red line. HKR all hit: the event has suspense, the post gives specific safegu...