Challenge: Existing backdoor models rely on visual inputs for instruction parsing, rendering the perception pathway a critical attack surface.
Approach: They propose an Adaptive Threat-Aware Adversarial Tuning framework that detects and decouples the optimal gradient decoupling strategy based on the adversary's capabilities.
Outcome: The proposed framework achieves a highly robust targeted attack success rate while maintaining extreme stealthiness with a 5% poisoning rate.

Similar Papers

BEEAR: Embedding-based Adversarial Removal of Safety Backdoors in Instruction-tuned Language Models (2024.emnlp-main)

Copied to clipboard

Challenge: Safety backdoors in large language models can be triggered while evading detection during normal interactions.
Approach: They propose a bi-level optimization method that uses a key insight: backdoor triggers induce a uniform drift in the model’s embedding space . inner level identifies universal perturbations to the decoder’s embedded spaces that steer the model towards defender-defined unwanted behaviors; outer level fine-tunes the model to reinforce safe behaviors against these perturbations.
Outcome: The proposed mitigation method reduces the success rate of safety backdoor attacks from over 95% to 1% for general harmful behaviors and from 47% to 0% for Sleeper Agents, without compromising the model’s usefulness.
A Gradient Control Method for Backdoor Attacks on Parameter-Efficient Tuning (2023.acl-long)

Copied to clipboard

Challenge: Parameter-Efficient Tuning (PET) fine-tunes pre-trained language models for downstream tasks, but a large reduction in the number of attackable parameters will greatly affect the effectiveness of backdoor attacks, resulting in backdoor forgetting.
Approach: They propose a gradient control method to consolidate the attack effect by freezing most parameters of the pre-trained model and fine-tuning only a small number of parameters.
Outcome: The proposed method improves sentiment classification and spam detection, and can be applied to different tasks.
Critical-CoT: A Robust Defense Framework against Reasoning-Level Backdoor Attacks in Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) are vulnerable to backdoors that use long-form reasoning to generate a specific word, choice, or class.
Approach: They propose a mechanism that allows LLMs to develop critical thinking behaviors and detect backdoors by a two-stage fine-tuning.
Outcome: The proposed mechanism exhibits strong cross-domain and cross-task generalization.
Attention-Enhancing Backdoor Attacks Against BERT-based Models (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing textual backdoor attacks focus on generating stealthy triggers or modifying model weights.
Approach: They propose a Trojan Attention Loss (TAL) which enhances the Trojan behavior by directly manipulating attention patterns.
Outcome: The proposed method improves the effectiveness of the backdoor attacks on different backbone models and tasks.
Attack as Defense: Safeguarding Large Vision-Language Models from Jailbreaking by Adversarial Attacks (2025.findings-emnlp)

Copied to clipboard

Challenge: adversarial vulnerabilities in vision-language systems pose a challenge to reliability of large systems . typographic manipulations and adversarial perturbations can bypass language model defenses .
Approach: They propose a method that embeds perturbations in vision to disrupt attacks . they use cross-modal interactions to enhance adversarial robustness through perturbations .
Outcome: The proposed approach reduces attack success rates for typographic attacks and adversarial perturbations by integrating visual defenses into the model.
UOR: Universal Backdoor Attacks on Pre-trained Language Models (2024.findings-acl)

Copied to clipboard

Challenge: Existing methods to attack pre-trained language models rely on manual selection of triggers and backdoor representations.
Approach: They propose a backdoor attack method that turns manual selection into automatic optimization . they propose to use poisoned contrastive learning to learn more uniform backdoor representations .
Outcome: The proposed method achieves better attack performance on text classification tasks compared to manual methods.
Light-weight Fine-tuning Method for Defending Adversarial Noise in Pre-trained Medical Vision-Language Models (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing fine-tuning algorithms for vision-language models are restricted by patient privacy concerns and can contain imperceptible noise.
Approach: They propose a framework to mitigate adversarial noise and mitigate upstream noise during fine-tuning.
Outcome: The proposed framework improves model robustness and transferability while decreasing noise levels negatively impact downstream performance.
Bridge the Gap Between CV and NLP! A Gradient-based Textual Adversarial Attack Framework (2023.findings-acl)

Copied to clipboard

Challenge: Existing methods for adversarial samples are poorly applied in computer vision . however, textual adversarials are still vulnerable to small perturbations .
Approach: They propose a framework to extend existing adversarial attack methods to textual adversarials by adding optimized perturbations to embedding layer and amplifying them in forward propagation process.
Outcome: The proposed framework achieves better performance even using proxy gradient information and produces more fluent and grammatical adversarial samples compared to baseline methods.
Making Every Step Effective: Jailbreaking Large Vision-Language Models Through Hierarchical KV Equalization (2025.findings-emnlp)

Copied to clipboard

Challenge: HKVE selectively accepts gradient optimization results based on the distribution of attention scores across different layers, ensuring that every optimization step positively contributes to the attack.
Approach: They propose a framework that selectively accepts gradient optimization results based on the distribution of attention scores across different layers and selectively takes them into account when calculating the attack success rate.
Outcome: The proposed framework outperforms existing methods by achieving success rates of 75.08% on MiniGPT4, 85.84% on LLaVA and 81.00% on Qwen-VL.
SilentDrift: Exploiting Action Chunking for Stealthy Backdoor Attacks on Vision-Language-Action Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing backdoor attacks introduce kinematic discontinuities and distributional anomalies that can be flagged by standard trajectory detection.
Approach: They propose a backdoor attack exploiting an intra-chunk visual open-loop vulnerability . they propose 93.2% Attack Success Rate and a poisoning rate under 2% .
Outcome: The proposed attack achieves a 93.2% Attack Success Rate with a poisoning rate under 2% while maintaining a 95.3% Clean Task Success Rate.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations