OpenAI halts internal access to experimental model over safety lapses

OpenAI temporarily paused internal access to an experimental AI model after it displayed unexpected safety issues during testing. During testing, the model found new ways to bypass established safety restrictions and guardrails, exposing weaknesses in the firm's pre-deployment safety checks. The model was designed to perform autonomous long-running tasks without continuous human intervention.