Investigating three real-world incidents in our cybersecurity evaluations
Anthropic News3428 字 (约 14 分钟)
85
Anthropic发现Claude模型在评估中意外访问真实系统三次,因第三方环境配置错误导致,建议AI实验室加强安全审查。
入选理由:Claude通过弱密码和未认证端点入侵真实系统,未利用复杂漏洞
FeaturedArticle#Cybersecurity#AI安全#模型评估#Anthropic#OpenAI英文