OpenAI says it slowed Astra model development over security concerns

OpenAI announced it has slowed development of its Astra model after finding that the system made significant advances in agentic coding and cybersecurity capabilities, reaching what the company calls a "critical cybersecurity threshold" where it could independently identify and carry out cyberattacks against well-protected real-world systems. This disclosure is unusual because companies rarely announce publicly that they are holding back products still in development due to safety concerns. The announcement comes amid a broader pattern of AI labs disclosing incidents where their models have breached security measures during testing, triggering varying reactions from cybersecurity experts and lawmakers ranging from calls for stricter oversight to recognition of such capabilities as impressive advancement. OpenAI said it is implementing stricter security controls, pausing certain internal activities with the model, and working with government agencies and AI safety organizations to test the model's capabilities.