Copertina

Asking Forever: Universal Activations Behind Turn Amplification in Conversational LLMs

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Coalson, Zachary, Fang, Bo, Hong, Sanghyun
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Cryptography and Security
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Discovering Universal Activation Directions for PII Leakage in Language Models
di: Marchyok, Leo, et al.
Pubblicazione: (2026)

IF-GUIDE: Influence Function-Guided Detoxification of LLMs
di: Coalson, Zachary, et al.
Pubblicazione: (2025)

Fail-Closed Alignment for Large Language Models
di: Coalson, Zachary, et al.
Pubblicazione: (2026)

Hard Work Does Not Always Pay Off: Poisoning Attacks on Neural Architecture Search
di: Coalson, Zachary, et al.
Pubblicazione: (2024)

PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flips
di: Coalson, Zachary, et al.
Pubblicazione: (2024)

Black-box Optimization of LLM Outputs by Asking for Directions
di: Zhang, Jie, et al.
Pubblicazione: (2025)

Turn-Based Structural Triggers: Prompt-Free Backdoors in Multi-Turn LLMs
di: Lu, Yiyang, et al.
Pubblicazione: (2026)

Mitigating Error Amplification in Fast Adversarial Training
di: Zhao, Mengnan, et al.
Pubblicazione: (2026)

Modeling Neural Networks with Privacy Using Neural Stochastic Differential Equations
di: Hong, Sanghyun, et al.
Pubblicazione: (2025)

Hessian-aware Training for Enhancing DNNs Resilience to Parameter Corruptions
di: Prato, Tahmid Hasan, et al.
Pubblicazione: (2025)

Unified Mechanism-Specific Amplification by Subsampling and Group Privacy Amplification
di: Schuchardt, Jan, et al.
Pubblicazione: (2024)

MADCAT: Combating Malware Detection Under Concept Drift with Test-Time Adaptation
di: Roh, Eunjin, et al.
Pubblicazione: (2025)

Privacy Amplification for Matrix Mechanisms
di: Choquette-Choo, Christopher A., et al.
Pubblicazione: (2023)

Privacy Backdoors: Enhancing Membership Inference through Poisoning Pre-trained Models
di: Wen, Yuxin, et al.
Pubblicazione: (2024)

Bias Amplification in RAG: Poisoning Knowledge Retrieval to Steer LLMs
di: Wang, Linlin, et al.
Pubblicazione: (2025)

Personalized Privacy Amplification via Importance Sampling
di: Fay, Dominik, et al.
Pubblicazione: (2023)

Leveraging Randomness in Model and Data Partitioning for Privacy Amplification
di: Dong, Andy, et al.
Pubblicazione: (2025)

Privacy Amplification for the Gaussian Mechanism via Bounded Support
di: Hu, Shengyuan, et al.
Pubblicazione: (2024)

Detecting Adversarial Data via Provable Adversarial Noise Amplification
di: Mumcu, Furkan, et al.
Pubblicazione: (2026)

Breaking Agents: Compromising Autonomous LLM Agents Through Malfunction Amplification
di: Zhang, Boyang, et al.
Pubblicazione: (2024)

Privacy Amplification Through Synthetic Data: Insights from Linear Regression
di: Pierquin, Clément, et al.
Pubblicazione: (2025)

Unveiling the Backdoor Mechanism Hidden Behind Catastrophic Overfitting in Fast Adversarial Training
di: Zhao, Mengnan, et al.
Pubblicazione: (2026)

Rényi Pufferfish Privacy: General Additive Noise Mechanisms and Privacy Amplification by Iteration
di: Pierquin, Clément, et al.
Pubblicazione: (2023)

Privacy Amplification by Structured Subsampling for Deep Differentially Private Time Series Forecasting
di: Schuchardt, Jan, et al.
Pubblicazione: (2025)

Turning Federated Learning Systems Into Covert Channels
di: Costa, Gabriele, et al.
Pubblicazione: (2021)

It's Our Loss: No Privacy Amplification for Hidden State DP-SGD With Non-Convex Loss
di: Annamalai, Meenatchi Sundaram Muthu Selva
Pubblicazione: (2024)

Certified Robustness to Clean-Label Poisoning Using Diffusion Denoising
di: Hong, Sanghyun, et al.
Pubblicazione: (2024)

Towards Strong Certified Defense with Universal Asymmetric Randomization
di: Hong, Hanbin, et al.
Pubblicazione: (2025)

Knowledge-Driven Multi-Turn Jailbreaking on Large Language Models
di: Li, Songze, et al.
Pubblicazione: (2026)

Universally Harmonizing Differential Privacy Mechanisms for Federated Learning: Boosting Accuracy and Convergence
di: Feng, Shuya, et al.
Pubblicazione: (2024)

AutoAdv: Automated Adversarial Prompting for Multi-Turn Jailbreaking of Large Language Models
di: Reddy, Aashray, et al.
Pubblicazione: (2025)

MetaCipher: A Time-Persistent and Universal Multi-Agent Framework for Cipher-Based Jailbreak Attacks for LLMs
di: Chen, Boyuan, et al.
Pubblicazione: (2025)

Shake to Leak: Fine-tuning Diffusion Models Can Amplify the Generative Privacy Risk
di: Li, Zhangheng, et al.
Pubblicazione: (2024)

Can LLMs be Fooled? Investigating Vulnerabilities in LLMs
di: Abdali, Sara, et al.
Pubblicazione: (2024)

LLMStinger: Jailbreaking LLMs using RL fine-tuned LLMs
di: Jha, Piyush, et al.
Pubblicazione: (2024)

HARP: Measuring Harm Amplification in Multi-Agent LLM Systems
di: Rahman, Md Hafizur, et al.
Pubblicazione: (2026)

Injecting Universal Jailbreak Backdoors into LLMs in Minutes
di: Chen, Zhuowei, et al.
Pubblicazione: (2025)

Compact: Approximating Complex Activation Functions for Secure Computation
di: Islam, Mazharul, et al.
Pubblicazione: (2023)

BAN: Detecting Backdoors Activated by Adversarial Neuron Noise
di: Xu, Xiaoyun, et al.
Pubblicazione: (2024)

xMLP: Revolutionizing Private Inference with Exclusive Square Activation
di: Li, Jiajie, et al.
Pubblicazione: (2024)