Skip to content
Longterm Wiki
Back

A Content-Based Framework for Cybersecurity Refusal Decisions (2025).

paper

Authors

Chuanhui Liu·Xiao Wang

Credibility Rating

3/5
Good(3)

Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.

Rating inherited from publication venue: arXiv

Note: The provided summary appears mismatched to this URL; this paper is about cybersecurity refusal policies for AI systems, relevant to those working on deployment safety, content moderation, and dual-use AI risks.

Paper Details

Citations
0
0 influential
Year
2025

Metadata

Importance: 55/100arxiv preprintprimary source

Abstract

This paper provides a comprehensive analysis of variational inference in latent variable models for survival analysis, emphasizing the distinctive challenges associated with applying variational methods to survival data. We identify a critical weakness in the existing methodology, demonstrating how a poorly designed variational distribution may hinder the objective of survival analysis tasks - modeling time-to-event distributions. We prove that the optimal variational distribution, which perfectly bounds the log-likelihood, may depend on the censoring mechanism. To address this issue, we propose censor-dependent variational inference (CDVI), tailored for latent variable models in survival analysis. More practically, we introduce CD-CVAE, a V-structure Variational Autoencoder (VAE) designed for the scalable implementation of CDVI. Further discussion extends some existing theories and training techniques to survival analysis. Extensive experiments validate our analysis and demonstrate significant improvements in the estimation of individual survival distributions.

Summary

This paper proposes a content-based framework for determining when AI systems should refuse cybersecurity-related requests, addressing the challenge of balancing security research utility against potential misuse. The framework provides principled criteria for refusal decisions based on the nature and specificity of requested content rather than topic alone.

Key Points

  • Proposes a structured framework for AI refusal decisions specifically tailored to cybersecurity requests
  • Argues refusal decisions should be content-based rather than topic-based, allowing legitimate security research while blocking harmful outputs
  • Addresses the dual-use dilemma in cybersecurity AI assistance where the same knowledge serves both attackers and defenders
  • Provides criteria to distinguish between educational/defensive content and operationally harmful cybersecurity assistance
  • Relevant to AI deployment policy and the design of safety measures in frontier models handling sensitive domains

Cited by 1 page

PageTypeQuality
Large Language ModelsCapability60.0

Cached Content Preview

HTTP 200Fetched Mar 15, 202698 KB
[2502.09591] Censor Dependent Variational Inference 
 
 
 
 
 
 
 
 
 
 
 

 
 
 

 
 
 
 
 
 
 Censor Dependent Variational Inference

 
 
 Chuanhui Liu
 
    
 Xiao Wang
 
 

 
 Abstract

 This paper provides a comprehensive analysis of variational inference in latent variable models for survival analysis, emphasizing the distinctive challenges associated with applying variational methods to survival data. We identify a critical weakness in the existing methodology, demonstrating how a poorly designed variational distribution may hinder the objective of survival analysis tasks—modeling time-to-event distributions. We prove that the optimal variational distribution, which perfectly bounds the log-likelihood, may depend on the censoring mechanism. To address this issue, we propose censor-dependent variational inference (CDVI), tailored for latent variable models in survival analysis. More practically, we introduce CD-CVAE, a V-structure Variational Autoencoder (VAE) designed for the scalable implementation of CDVI. Further discussion extends some existing theories and training techniques to survival analysis. Extensive experiments validate our analysis and demonstrate significant improvements in the estimation of individual survival distributions. Codes can be found at https://github.com/ChuanhuiLiu/CDVI .

 
 Machine Learning, ICML
 
 
 
 
 
 \printAffiliations 
 
 
 
 1 Introduction

 
 Survival analysis, a fundamental topic in statistics, finds wide-ranging applications across healthcare, insurance, quality management, and finance. It focuses on modeling the relationship between time-to-event outcomes and individual demographic covariates, where the event of interest could be death, disease progression, or similar occurrences. A key challenge in survival analysis arises from censored observations, which provide only partial information about the survival time, necessitating specialized methods to handle such data effectively.

 
 
 Deep learning has emerged as a powerful paradigm to advance survival analysis (Wiegrebe et al., 2024 ) . Recent studies focus on modeling time-to-event distributions via latent variable survival models (LVSMs), applying various probabilistic assumptions and inference techniques. For example, Ranganath et al. ( 2016 ) assumed that the prior of Z Z belongs to the class of deep exponential family distributions (Brown, 1986 ) . Instead, deep survival machine (Nagpal et al., 2021a ) considered the finite discrete latent space, and the time-to-event distribution is one of the finite Gumbel or normal distributions. For discrete time-to-event, (Xiu et al., 2020 ) modeled a softmax-activated neural network incorporating the Nelson-Aalen estimator (Aalen, 1978 ) , while Apellániz et al. ( 2024 ) followed a similar setup, developing variational autoencoders (VAEs) (Kingma & Welling, 2014 ; Rezende et al., 2014 ) for continuous time-to-event. These new advances of LVSM have demonstrated superior performance across various metri

... (truncated, 98 KB total)
Resource ID: 8fac188a634d15e6 | Stable ID: ZDBiMjQ1Nm