A Unified Optimization Framework for Backdoor Attacks in Large Language Models

doi:10.1016/j.inffus.2026.104221

A Unified Optimization Framework for Backdoor Attacks in Large Language Models

2026-02-12

0

PRE

AI

DOI：10.1016/j.inffus.2026.104221

Original

Original request for help

Save

Abstract

En 中文

• Propose a framework for backdoor attacks in large language models. • Theoretically prove the sufficiency of partial parameter updates. • Resolve gradient conflicts between clean and adversarial objectives. • Validate higher success rates with minimal clean-task performance drop.

Keywords:

Backdoor Attacks

Large Language Models

Optimization Framework

Gradient Conflict

Parameter Updates

A Unified Optimization Framework for Backdoor Attacks in Large Language Models

Abstract

Journal

Organization