OpenAI因前沿网络安全风险暂缓强化学习训练

19.08.2026 ForkLog OpenAI has temporarily slowed the scaling of new AI models and paused reinforcement learning (RL) training on its latest systems intended for deployment for two weeks. > As models become more capable, the risks associated with developing and testing them internally also grow. > > We temporarily paused reinforcement learning (RL) training on our latest models intended for deployment for two weeks while we hardened and red-teamed our research… > > — OpenAI (@OpenAI) August 18, 2026 Two developments influenced the decision: the Hugging Face incident and a preliminary assessment of Astra, after which OpenAI said it could not rule out the model reaching Critical — the highest level of cyber capabilities in the company’s Preparedness Framework. [...] In parallel, OpenAI paused RL training on its latest models intended for deployment for two weeks. It used this time to harden research environments, conduct red-teaming, and expand monitoring. The largest planned RL run h

As models become more capable, the risks associated with developing and testing them internally also grow. Our standards for monitoring, alignment, and security must stay ahead of those risks. We wanted to take the time necessary to meet those standards, so we temporarily slowed the pace of scaling. This included a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment while we further hardened and red-teamed our research environments and expanded the coverage of our monitoring systems. Our largest planned frontier RL run remains on hold while we conduct smaller-scale training and evaluations to assess model behavior, validate our safeguards, and establish more evidence of alignment before proceeding. [...] What’s next Strengthening safeguards for more capable models Securing our research environments Expanding chain-of-thought monitoring Advancing alignment research What’s next Over the past several weeks, two

OpenAI said: > "As models become more capable, the risks associated with developing and testing them internally also grow. Our standards for monitoring, alignment, and security must stay ahead of those risks. We wanted to take the time necessary to meet those standards, so we temporarily slowed the pace of scaling. This included a two-week pause in reinforcement learning (RL) training on our latest models intended for deployment while we further hardened and red-teamed our research environments and expanded the coverage of our monitoring systems. Our largest planned frontier RL run remains on hold while we conduct smaller-scale training and evaluations to assess model behavior, validate our safeguards, and establish more evidence of alignment before proceeding." [...] The extra time will be used to ensure AI systems behave as intended and under human oversight. OpenAI said its development for more capable models, the ones that will likely be used for cybersecurity defenses, will have

The company stated: “As models become more capable, the risks associated with developing and testing them internally also grow. Our

As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforcement

@OpenAI OpenAI @OpenAI As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforcement learning (RL) training on our latest models intended for deployment for two weeks while we hardened and red-teamed our research environments and expanded monitoring coverage. Our largest planned frontier RL run remains on hold while smaller-scale training and evaluations validate these safeguards and establish more evidence of alignment. openai.com/index/pacing-m… Pacing model development in an era of cyber-critical capabilities Pacing model development in an era of cyber-critical capabilitiesFrom openai.com 6:13 PM · Aug 18, 20261.8MViews 646 @OpenAI OpenAI @OpenAI [...] 646 @OpenAI OpenAI @OpenAI As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforcement learning (RL) training on our latest models intended for de