OpenAI pauses frontier RL training until safety standards catch up
- Source
- x.com
- Date
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 capabilities were outstripping the pace of safety and alignment. We care very deeply about AI safety. We believe the entire field will have to coordinate on shared safety standards, but will act unilaterally in the meantime. We expect confidence in safety to increasingly set the pace of AI progress. We are optimistic about the alignment work we are doing, and we remain committed to making frontier capabilities widely available. https://t.co/51kvKfbfrO
(We still expect to ship great new models soon; this impacts further-out releases.)
Context
OpenAI's August 18, 2026 account describes a pause in some model training while it strengthened research safeguards. It reports a two-week interruption to reinforcement learning, training that improves behavior through feedback, and says its largest planned frontier run remained on hold while smaller tests continued. In his post, CEO Sam Altman said new models were still expected soon and that further-out releases were affected.
The account describes strengthening safeguards around the company's research systems, including limits on code execution and network access and expanded monitoring. This followed the earlier delay to Astra's wider release, so the company was changing the conditions for developing and testing powerful models as well as deciding who could use them.
- alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.
Checking sign-in…
Loading comments…