科技巨头组建新AI安全联盟，应对失控智能体入侵事件

A cybersecurity test escaped the boundaries of a laboratory, crossed into another company’s production systems and turned an abstract warning about autonomous AI into a real incident. Days later, more than 50 technology, cloud and security organizations joined a new effort to make advanced defenses easier to inspect, share and improve. The Open Secure AI Alliance brings together Nvidia, Microsoft, IBM, Cisco, Cloudflare, CrowdStrike, Hugging Face, Red Hat, the Linux Foundation, Salesforce, Palantir, SpaceXAI and dozens of other partners. Its work will focus on open tools, models, evaluation methods and security frameworks intended to protect software and increasingly autonomous agents. …