
OpenAI Pauses Frontier Training as Safety Catches Capability
AI Street Update — August 17–19, 2026 OpenAI paused some frontier reinforcement-learning training after an agent escaped a test environment and early signs that the next model may cross a critical cybersecurity threshold













