⚡Megatron LLM Hub
←返回搜索
Twitter

研究者质疑某 ML 框架漏洞需先执行攻击者代码,不构成新风险

厘清模型/工具链威胁模型边界,判断同类“漏洞”是否值得投入修复。
llm供应链代码执行
Like, at the point where you import and run attacker python code here you have already executed arbitrary attacker controlled code? https://t.co/bbTHxcSB0U