Skip to content

Translation in progress Some entries still include text in the original language.

Oct 12 · Monday

Defenses and guardrails · Papers

Found 2 papers
Filter1

Attack type

Jailbreaks61Prompt injection9Model tampering2
Select any value within a category; different categories must all match. Counts include the other filters you have selected.
Attack type：Model tamperingTopic: Defenses and guardrailsClear all
  1. InterpretabilityarXiv 2609.06934

    论文提出拒答的几何解释：事后安全训练脆弱，预训练期持续安全共训更抗攻击

    解释后验安全的权重几何脆弱性，并提出预训练持续安全共训练

    arXiv
    2609.06934
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen-2.5-7B, Qwen-2.5-7B-Instruct, Llama-3-8B and others15

    Abstract论文将后验安全脆弱性归因于正交抑制，并提出通过预训练持续共训练实现具备微调抗性的安全防护。

    Paper analysisRead analysis
  2. DefensesarXiv 2609.02293

    SEAL：通过共享专家对齐强化 MoE 全局安全

    提出 SEAL，在 MoE 共享专家上做 DPO-LoRA 对齐以抵御越狱与权重篡改

    arXiv
    2609.02293
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Attacker
    White-box, Black-box
    Evaluated models
    Qwen1.5-MoE-A2.7B, DeepSeekMoE-16B, GLM-4.7-Flash and others4
    AbstractSEAL 把 DPO-LoRA 限制在混合 MoE 共享专家上，SEAL++再保护已有安全神经元方向。

    SEAL是面向混合 MoE 的训练时防御：只对齐始终执行的共享专家，使安全不依赖路由器是否把 token 送到安全专家。

    Paper analysisRead analysis
You have reached the end