人类开发实验性AI系统Mythos以发现软件漏洞,但由于滥用问题限制了访问. Anthropic develops experimental AI system Mythos to find software vulnerabilities but restricts access due to misuse concerns.

开发一种名为"神话"的实验性人工智能系统, Anthropic is developing an experimental AI system called Mythos designed to identify high-severity software vulnerabilities. 虽然独立测试显示它在73%的时间里成功完成了专家级网络安全任务, 但它的强大功能引发了对潜在滥用的严重担忧. While independent tests show it succeeded in expert-level cybersecurity tasks about 73% of the time, its powerful capabilities raise serious concerns about potential misuse. 系统目前还没有公开, 只在一小部分主要科技公司进行测试, 并受到严格控制以防止滥用. Consequently, the system is not publicly available and is being tested only with a select group of major tech companies under strict controls to prevent abuse.