OpenAI pauses some frontier training to strengthen security and monitoring
geoffreyirving · x · 2026-08-19
Addressing reports that OpenAI is dedicating 20% of research inference compute to chain-of-thought monitoring, OpenAI's Merett confirmed that the team has temporarily slowed some frontier training to bolster security. The largest planned frontier RL run remains on hold while smaller-scale training helps test safeguards. Merett stated that confidence in safety will increasingly set the pace of AI development and emphasized the urgent need for coordination tools.
More from Companies & People
- Researcher mhrezaeics joins Mira Murati's Thinking Machines Lab — simonguozirui · 2026-10-16
- Apple and Meta reportedly collaborating under new Apple CEO Ternus, signaling thaw — sudoraohacker · 2026-10-16
- PE operating partner shares real agent deployments: 59/60 accuracy at $250/month — alex_verem · 2026-10-16
- Bittensor's DecisionModels SN92 launch hits 914K views, out-tweeting Meta — markjeffrey · 2026-10-16
- Linear and Figma founders poke fun as UI thrives despite 'TUI is the future' predictions — zeeg · 2026-10-16
- Mustafa Suleyman's 'x% of FLOPS' boast mocked as Microsoft sits on ~12GW of compute — teortaxesTex · 2026-10-16