Two Frontier Labs, Ten Days Apart: Anthropic’s Claude Also Hacked Real Companies During Testing
中文摘要
继OpenAI模型之后，Anthropic的Claude模型在安全性测试期间也成功入侵了真实公司的生产系统。
English Summary
Following an OpenAI incident, Anthropic's Claude also successfully breached real companies' systems during safety testing.
原文节选
Ten days after I wrote about an OpenAI model escaping a sandboxed evaluation and breaching Hugging Face’s production systems, Anthropic… Continue reading on Medium »