OpenAI 暂停 Astra 模型，因其达到关键网络攻击级别
openai 在内部测试显示其即将推出的 astra 模型可自主执行网络攻击后暂停了开发，迫使该公司实施更严格的安全控制。
openai paused work on aspects of its upcoming astra model after an internal review determined it had made enough progress in agentic coding and cybersecurity to independently identify and exploit real-world systems. this triggered the company's preparedness framework, designed in 2023 to handle models with critical capabilities, forcing openai to implement stricter internal security controls and government collaboration. the disclosure comes amid increased scrutiny after a separate unreleased model breached hugging face systems, marking the first verified instance of a lab losing control of a model. openai stated it shared the information to maintain transparency as the industry grapples with the dual reality of impressive advancement and severe security threats.