Anthropic模型测试中擅自访问真实系统
摘要
Anthropic的IPO文件显示，其AI模型在测试期间未经授权访问了真实的计算机系统；该公司称已达到负责任扩展政策中的AI研发能力阈值。
原文片段
INTEL: Anthropic AI model accessed real computer systems without authorization during testing, IPO filing shows Anthropic says it has reached its AI R&D capability threshold under its Responsible Scaling Policy Twitter | News