未发布先警告 OpenAI：新AI模型或构成高级别网络安全风险

Core Viewpoint - OpenAI warns that its upcoming AI models may pose "high-level" cybersecurity risks due to their rapid performance improvements [1] Group 1: AI Model Risks - The new AI models could autonomously develop zero-day exploits capable of attacking well-protected systems or assist in complex corporate or industrial intrusion operations [1] - OpenAI emphasizes the need for enhanced defensive capabilities in its models to mitigate these risks [1] Group 2: Defensive Measures - The company is investing resources to strengthen the models' abilities in defensive cybersecurity tasks and to develop tools that facilitate code auditing and vulnerability remediation for defenders [1] - OpenAI plans to implement a combination of measures, including access control, infrastructure hardening, export traffic control, and monitoring to address cybersecurity risks [1] Group 3: Initiatives and Collaborations - OpenAI will initiate a program to explore tiered access for qualified users and customers engaged in cybersecurity defense to utilize enhanced features [1] - The establishment of a "Frontier Risk Council" will involve collaboration with experienced cybersecurity experts and practitioners, initially focusing on cybersecurity before expanding to other advanced capability areas [1]