OpenAI is slowing down its next model over ‘critical’ cyber risk

OpenAI has paused internal development of its upcoming "Astra" model after testing showed it may have reached "Critical" cyber capabilities, meaning it could potentially hack hardened systems or plan complex attacks without human help. No previous OpenAI model had reached this level. The company is isolating test environments and bringing in government agencies to help evaluate the model, though critics note that AI containment has already failed multiple times recently.