Tag
The article argues that gating smart models like Mythos reduces societal safety, advocating for wider distribution of AI technology to secure the vast ecosystem of open-source and closed-source software projects.
This paper proposes methods for protecting large language models against unauthorized knowledge distillation by rewriting reasoning traces to degrade training usefulness while preserving correctness, and embedding verifiable watermarks in distilled student models. The approach uses instruction-based and gradient-based rewriting techniques to achieve anti-distillation effects without compromising teacher model performance.
This paper demonstrates that deep neural networks are catastrophically vulnerable to minimal sign-bit flips in parameters, introducing DNL and 1P-DNL methods to identify critical vulnerable parameters without data or optimization. The vulnerability spans multiple domains including image classification, object detection, instance segmentation, and language models, with practical implications for model security.
Anthropic announced new defenses against illicit distillation attacks on Claude, claiming a 90-99% reduction in harvested data, while warning that distilled models like Zhipu's GLM-5.3 pose cybersecurity and national security risks.
Anthropic宣布在抵御针对Claude的蒸馏攻击方面取得重大进展,早期数据显示已将蒸馏者窃取的数据量减少约90%至99%。文章将非法蒸馏描述为国家安全风险,指出Zhipu在开发GLM-5.3时曾试图蒸馏Claude的网络攻击能力,且部分蒸馏模型的能力已被整合进外国军事和情报系统。