OpenAI CEO Sam Altman said that his company is pausing reinforcement learning (RL) of frontier AI models, including the upcoming Astra model, over misalignment and cybersecurity risks, after its AI models breached other organizations.

In a post on X on Wednesday, Altman said that OpenAI is pausing RL training for some models until the researchers develop “appropriate alignment, security and monitoring standards.”

RL Training refers to a branch of machine learning where the AI model learns through activities including trial-and-error methods. The statement by OpenAI comes a day after Anthropic revised its assessment of the risk from misaligned AI behaviour from “very low” to “low,” citing increased uncertainty following recent disclosures related to model behaviour in cybersecurity evaluations.

“Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment,” wrote Altman on X.

We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model…

— Sam Altman (@sama) August 18, 2026

OpenAI has claimed that two recent incidents involving frontier AI models of the company has proved that AI’s ability to learn and adapt has outpaced the lab’s efforts to grow safeguards around them. The company has claimed to put a two week pause on AI models’ training.

One such incident was the hacking of Hugging Face platform in July this year when two models of OpenAI had escaped sandbox and breached an open-source platform.

According to OpenAI, another incident was when they put their upcoming AI model Astra in ‘critical cybersecurity capability.’

“ 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,” read a statement from OpenAI.

The AI company has also claimed to put stringent safeguards around their frontier AI models including stronger isolation or “sandboxes” for models, isolating models from internet or network access and enhanced security tests.

“While some Astra training and evaluations meet those requirements, a significant number of workloads remain paused until they are fully migrated and enhanced to meet the new security bar. We are prioritizing safety and alignment workloads for migration to these new environments first,” said OpenAI.

Also Read: OpenAI Slows Down Astra Launch, Flags Risks Over Critical Cyber Capabilities