Skip to content

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

Oct 12 · Monday

Defenses and guardrails · Papers

Found 3 papers
Filter2

Defense type

Model hardening3

Impact

Harmful content3

System type

Models and APIs3

Layer

Prompt layer3Model layer6Training and data layer3

System component

Multimodal3
Select any value within a category; different categories must all match. Counts include the other filters you have selected.
Layer：Training and data layerSystem component：MultimodalTopic: Defenses and guardrailsClear all
  1. DefensesarXiv 2610.05637

    Study shows VLM visual grounding refusal rates are on average 31 to 59 percentage points lower by harm domain

    提出融合定位拒答与能力保留的安全微调，补齐视觉定位输出的拒答缺口

    arXiv
    2610.05637
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Claude Sonnet 4.6, Gemini 3.5 Flash, Molmo2-8B and others5

    Abstract论文揭示了 VLM 定位安全落后于 VQA 的对齐断层，提出三成分微调方案有效弥合差距并保留定位能力，表征分析揭示了对齐机制。

    Paper analysisRead analysis
  2. DefensesarXiv 2610.05401

    AVCE：面向扩散模型可验证概念擦除的审计感知遗忘框架

    提出审计感知遗忘框架 AVCE，从文本与视觉通路擦除扩散模型概念

    arXiv
    2610.05401
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Attacker
    White-box, Black-box
    Evaluated models
    Stable Diffusion XL (SDXL), Stable Diffusion v1.5 (SD v1.5), Flux 1.0
    AbstractAVCE 在审计锚点上改两条注意力通路。

    AVCE 是面向扩散模型的概念擦除方法，要让擦除经得起绕过文本条件的潜空间审计，而不只是挡住对抗提示词。

    Paper analysisRead analysis
  3. DefensesarXiv 2609.37837

    Intent-Privilege OPSD：让视觉语言模型在隐式风险下兼顾安全与有用

    提出 Intent-Privilege OPSD，对齐视觉语言模型在隐式风险下的安全与帮助性

    arXiv
    2609.37837
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-VL-4B-Instruct, LLaVA, LLaVA-1.5-7B and others4
    Abstract作者报告训练期证据化意图蒸馏兼顾安全与帮助性，效果受标注质量和评测设置影响。

    Intent-Privilege OPSD 是针对跨模态隐式风险的训练期安全—帮助性对齐方法。作者关注的不是直接拒绝敏感请求，而是按可观察图文证据保留正常帮助、约束有害信息。

    Paper analysisRead analysis
You have reached the end