Story perspectives

OpenAI Halts Reinforcement Learning for Two Weeks After Incident

8/19/2026

Summary

Story summary

OpenAI announced in the United States on Tuesday, August 18 2026 a two-week halt of reinforcement learning for its models after the Hugging Face incident.
VP of research Amelia Glaese said pause adds alerts within 30 minutes, adds 20 % compute, and will enforce controls on largest models.
Chief scientist Jakob Pachocki said OpenAI has informed the White House under Preparedness Framework because self-imposed pause may let Astra cross the “Critical” cybersecurity capability threshold.

1 of 1

Story summary

OpenAI announced in the United States on Tuesday, August 18 2026 a two-week halt of reinforcement learning for its models after the Hugging Face incident.
VP of research Amelia Glaese said pause adds alerts within 30 minutes, adds 20 % compute, and will enforce controls on largest models.
Chief scientist Jakob Pachocki said OpenAI has informed the White House under Preparedness Framework because self-imposed pause may let Astra cross the “Critical” cybersecurity capability threshold.