ACL 2026 — Trust & Safety | Rajat Kanti Bhattacharjee

Trust & Safety

181 papers Links not yet available — ACL proceedings pending on ACL Anthology.

MauBERT: Universal Phonetic Inductive Biases for Few-Shot Acoustic Units Discovery
Reasoning over Precedents Alongside Statutes: Case-Augmented Deliberative Alignment for LLM Safety
Efficient Provably Secure Linguistic Steganography via Range Coding
Anchored Sliding Window: Toward Robust and Imperceptible Linguistic Steganography
SPAGBias: Uncovering and Tracing Structured Spatial Gender Bias in Large Language Models
Identifying Bias in Machine-generated Text Detection
Toward Secure Tuning: Mitigating Security Risks from Instruction Fine-Tuning
Beyond Surface-Level Detection: Towards Cognitive-Driven Defense Against Jailbreak Attacks via Meta-Operations Reasoning
Bias Fitting to Mitigate Length Bias of Reward Model in RLHF
Debiased Orthogonal Boundary-Driven Efficient Noise Mitigation
APPSI-139: A Parallel Corpus of English Application Privacy Policy Summarization and Interpretation
Between a Rock and a Hard Place: The Tension Between Ethical Reasoning and Safety Alignment in LLMs
AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns
Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation
Towards Order Fairness: Mitigating LLMs Order Sensitivity through Dual Group Advantage Optimization
PII-Bench: Evaluating Query-Aware Privacy Protection Systems
Reasoning Structure Matters for Safety Alignment of Reasoning Models
Retrievals Can Be Detrimental: Unveiling the Backdoor Vulnerability of Retrieval-Augmented Diffusion Models
Common to Whom? Regional Cultural Commonsense and LLM Bias in India
LogicPoison: Logical Attacks on Graph Retrieval-Augmented Generation
Compete to Complete: Co-opetition Adversarial Learning for Retrieval-Augmented Generation
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
ContextLens: Modeling Imperfect Privacy and Safety Context for Legal Compliance
QuantileMark: A Message-Symmetric Multi-bit Watermark for LLMs
Empirical Analysis of Decoding Biases in Masked Diffusion Models
On the Hidden Objective Biases of Group-based Reinforcement Learning
SharedRequest: Privacy-Preserving Model‑Agnostic Inference for Large Language Models
Evo-Attacker: Memory-Augmented Reinforcement Learning for Long-Horizon Tool Attacks on LLM-MAS
All Languages Matter: Understanding and Mitigating Language Bias in Multilingual RAG
Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level Diagnosis
Parity-Aware Byte-Pair Encoding: Improving Cross-lingual Fairness in Tokenization
JPU: Bridging Jailbreak Defense and Unlearning via On-Policy Path Rectification
Enhancing the Transferability of Jailbreak Attacks on Large Language Models via Exploiting Reparameterization Invariance
GAMBIT: A Gamified Jailbreak Framework for Multimodal Large Language Models
Detecting RAG Extraction Attack via Dual-Path Runtime Integrity Game
MirageBackdoor: A Stealthy Attack that Induces Think-Well-Answer-Wrong Reasoning
Privacy Collapse: Benign Fine-Tuning Can Break Contextual Privacy in Language Models
Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling
OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows
SGT: Securing Open-Source LLMs Against Malicious Fine-tuning via Safety Guidance Trigger
GKnow: Measuring the Entanglement of Gender Bias and Factual Gender
Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images
Critical-CoT: A Robust Defense Framework against Reasoning-Level Backdoor Attacks in Large Language Models
Defenses Against Prompt Attacks Learn Surface Heuristics
Towards Mitigating Modality Bias in Vision-Language Models for Temporal Action Localization
DMN: A Compositional Framework for Jailbreaking Multimodal LLMs with Multi-Image Inputs
XOXO: Stealthy Cross-Origin Context Poisoning Attacks against AI Coding Assistants
Provably Safe Offline-to-Online RL: Decoupling Learning from Data-Driven Safety Enforcement
SHAPE: Unifying Safety, Helpfulness and Pedagogy for Educational LLMs
AgentMark: Utility-Preserving Behavioral Watermarking for Agents
Stereotype Bias in a Bilingual Setting: A Culturally Grounded Evaluation in Kazakhstan
When Identity Skews Debate: Anonymization for Bias-Reduced Multi-Agent Reasoning
XMark: Reliable Multi-Bit Watermarking for LLM-Generated Texts
Attention Under Attack: Analog Noise Effects and Mechanistic Vulnerabilities in Transformer Models
Protecting Bystander Privacy via Selective Hearing in Audio LLMs
To Lie or Not to Lie? Investigating The Biased Spread of Global Lies by LLMs
When Vision-Language Models Judge Without Seeing: Exposing Informativeness Bias
VIGNETTE: Socially Grounded Bias Evaluation for Vision-Language Models
Query-Efficient Agentic Graph Extraction Attacks on GraphRAG Systems
CoLA: A Choice Leakage Attack Framework to Expose Privacy Risks in Subset Training
Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets
Quantifying Metric and Model Agreement in Bias Evaluation of Large Language Models
Privacy-preserving Prosody Representation Learning
Activation-Guided Local Editing for Jailbreaking Attacks
Controlling Distributional Bias in Multi-Round LLM Generation via KL-Optimized Fine-Tuning
Reference Attack: A New Cross-Modal Jailbreaking Attack against Multimodal Large Language Models
Persona-Grounded Safety Evaluation of AI Companions in Multi-Turn Conversations
Seeing No Evil: Blinding Large Vision-Language Models to Safety Instructions via Adversarial Attention Hijacking
Dynamics of Cognitive Heterogeneity: Investigating Behavioral Biases in Multi-Stage Supply Chains with LLM-Based Simulation
Knowledge Poisoning Attacks on Medical Multi-Modal Retrieval-Augmented Generation
When Bigger Isn’t Better: A Comprehensive Fairness Evaluation of Political Bias in Multi-News Summarisation
NaturalSloth: Revisiting Denial-of-Service Attacks on Large Language Models
Preconditioned Test-Time Adaptation for Out-of-Distribution Debiasing in Narrative Generation
Backdoor Collapse: Eliminating Unknown Threats Via Known Backdoor Aggregation In Language Models
How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study
Probing the Safety Robustness of LLMs in Latent Space
USB: A COMPREHENSIVE AND UNIFIED SAFETY EVALUATION BENCHMARK FOR MULTIMODAL LARGE LANGUAGE MODELS
Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring
Benchmarking Web Agent Safety under E-commerce Deceptive Interfaces
CheckMIABench: Firm Foundations For Membership Inference Attacks on Language Models
Debiasing Reward Models via Causally Motivated Inference-Time Intervention
Steganography Beyond Pixels: Reimagining Image Steganography as Cross-Modal Linguistic Communication
Agent-based Substructure Counting under Local Differential Privacy
DEBAR: Mitigating Contextual Bias in Cross-Document Relation Extraction via Dual-Stream Decoupling
InsideOut: Measuring and Mitigating Insider–Outsider Bias in Interview Script Generation
Merging Triggers, Breaking Backdoors: Defensive Poisoning for Instruction-Tuned Language Models
Safe-FedLLM: Delving into the Safety of Federated Large Language Models
When Efficiency Meets Safety: A Benchmark Security Analysis of KV Cache Compression in Large Language Models
Can Persona-Prompted LLMs Emulate Subgroup Values? An Empirical Analysis of Generalisability and Fairness in Cultural Alignment
Into the Gray Zone: Domain Contexts Can Blur LLM Safety Boundaries
PRISP: Privacy-Safe Few-Shot Personalization via Lightweight Adaptation
MM-JudgeBias: A Benchmark for Evaluating Compositional Biases in MLLM-as-a-Judge
SafetyMem: Adaptive Jailbreak Defense via Dual-Component Safety Memory
CrossGuard: Safeguarding MLLMs against Joint-Modal Implicit Malicious Attacks
Jailbreaking Multimodal Large Language Models using Multi-Clip Video
Towards Privacy-Preserving Large Language Model: Text-free Inference Through Alignment and Adaptation
Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors
Inhibitory Attacks on Backdoor-based Fingerprinting for Large Language Models
TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment
Audio Jailbreak: An Open Comprehensive Benchmark for Jailbreaking Large Audio-Language Models
When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue Agents
SoundBreak: A Systematic Study of Audio-Only Adversarial Attacks on Trimodal Models
You Can Have a Second Chance: Unbiased and Multi-bit Watermarking for Diffusion Language Models with Regret-based Remasking
OASIS: Mitigating Harmful Fine-tuning Attacks on LLMs via Orthogonal and Adaptive Safety Alignment Strategy
Fair-CCD: Mitigating Bias in Large Language Models for Tabular Classification Through Context-Contrastive Decoding
Selective Test-Time Debiasing for CLIP via Reward Gating
EIFFEL: a novel benchmark to measure bias of English heavy training on French idiomatic expressions
N-GLARE: An Non-Generative Latent Representation-Efficient LLM Safety Evaluator
On the (In-)Security of the Shuffling Defense in the Transformer Secure Inference
Location Not Found: Exposing Implicit Local and Global Biases in Multilingual LLMs
Detecting What Queries Seek: Steering LLM Safety with FFN Output Activation Monitoring
Beyond Explicit Refusals: Soft-Failure Attacks on Retrieval-Augmented Generation
Understanding and Mitigating Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks
Evaluating Answer Leakage Robustness of LLM Tutors against Adversarial Student Attacks
Inverting the Shield: Systematically Generating Safety Tests from Policy Specifications
The “Knowledge–Behavior Gap” in Cultural Taboo Safety of Large Language Models
TAMAS: Benchmarking Adversarial Risks in Multi-Agent LLM Systems
ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments
SAFO: Stable Adaptive Fairness Optimization for LLM-Based Social Survey Simulation
Do LLMs Know Tool Irrelevance? Demystifying Structural Alignment Bias in Tool Invocations
Backdoors in RLVR: Jailbreak Backdoors in LLMs From Verifiable Reward
Defense Against Knowledge Poisoning Attack on GraphRAG
MM-PoisonRAG: Disrupting Multimodal RAG with Local and Global Knowledge Poisoning Attacks
Observations and Remedies for Large Language Model Bias in Self-Consuming Performative Loop
Resolving the Security-Auditability Dilemma with Auditable Latent Chain-of-Thought Alignment
More Thinking, Less Talking: Internalizing Deliberative Safety into LLM Parameters
Conjunctive Prompt Attacks in Multi-Agent LLM Systems
SMARTER: A Data-efficient Framework to Improve Toxicity Detection with Explanation via Self-augmenting Large Language Models
Confident, Calibrated, or Complicit: Safety Alignment and Ideological Bias in LLM Hate Speech Detection
New Terms, New Toxicity: Consensus-based Chinese Neologism Toxicity Detection via Search-Augmented LLMs
Can LLM Safety Be Ensured by Constraining Parameter Regions?
Mitigating Selection Bias in Large Language Models via Permutation-Aware GRPO
SAME: Safety-Aware Model Editing Guided by Safety Transformation
ARF-RLHF: Adaptive Reward-Following for RLHF through Emotion-Driven Self-Supervision and Trace-Biased Dynamic Optimization
User Perceptions vs. Proxy LLM Judges: Privacy and Helpfulness in LLM Responses to Privacy-Sensitive Scenarios
Challenging the Explanation Based on Preceding Tokens: Discovering Transferable Non-Literal Biasing
SWAN: Semantic Watermarking with Abstract Meaning Representation
Framing Political Bias in Multilingual LLMs Across Pakistani Languages
SSG: Logit-Balanced Vocabulary Partitioning for LLM Watermarking
The Side Effects of Being Smart: Safety Risks in MLLMs’ Multi-Image Reasoning
From TDMA to CDMA: A Multi-bit Watermark for Diffusion Language Models
FairQE: Multi-Agent Framework for Mitigating Gender Bias in Translation Quality Estimation
When Background Matters: Breaking Medical Vision Language Models by Transferable Attack
When Efficiency Becomes a Vulnerability: Computational Cost Attacks on WebAgents
BEFT: Bias-Efficient Fine-Tuning of Language Models in Low-Data Regimes
BlindGuard: Safeguarding LLM-based Multi-Agent Systems under Unknown Attacks
Black-Box Membership Inference Attacks for Video Training Data in Multimodal Large Language Models
Mitigating Safety Context Amnesia in Multimodal Reasoning Models via Intent-Guided Safety Reasoning
Robust Membership Inference for Large Language Models under Adversarial Generative Corruption
DARM: Distribution-Aware Reward Modeling by Alleviating Biases from Low Preference-Context Dependency Data
ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs
LLM Safety From Within: Detecting Harmful Content with Internal Representations
Measuring Social Bias in Vision-Language Models with Face-Only Counterfactuals from Real Photos
HarDBench: A Benchmark for Draft-Based Co-Authoring Jailbreak Attacks for Safe Human–LLM Collaborative Writing
Retrieval-Augmented Defense: Adaptive and Controllable Jailbreak Prevention for Large Language Models
LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety
Multimodal Safety Evaluation in Generative Agent Social Simulations
SafeMT: Multi-turn Safety for Multimodal Language Models
Estimating the Black-box LLM Uncertainty with Distribution-Aligned Adversarial Distillation
AutoRAN: Automated Hijacking of Safety Reasoning in Large Reasoning Models
Simple Agents, Biased Judges: Efficient Multi-Party Dialogue Generation & The Evaluation Gap
Deep Research with Open-Domain Evaluation and Multi-Stage Guardrails for Safety
FAIRGAMER: Evaluating Social Biases in LLM-Based Video Game NPCs
A Lightweight Explainable Guardrail for Prompt Safety
Activation Decomposition and Steering for LLM Backdoor Remediation
Investigating Counterfactual Unfairness in LLMs towards Identities through Humor
Route to Rome Attack: Directing LLM Routers to Expensive Models via Adversarial Suffix Optimization
Don’t Corrupt the Fact: A Trustworthy RAG Watermarking Framework based on Dual Factual Shield
Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration
Adversarial Metric Learning for Fine-Grained Emotion Classification
SHARP: Self-adaptive Harmful Category-aware Prompt Generation for Black-box Jailbreaking
JARVIS or Ultron? A Survey on the Safety and Security Threats of Computer-Using Agents
Evaluating Structure-Aware Retrieval and Safety in Statute-Centric Legal QA
A Linguistics-Aware LLM Watermarking via Syntactic Predictability
Privacy-R1: Privacy-Aware Multi-LLM Agent Collaboration via Reinforcement Learning
Jailbreak-Zero: A Path to Pareto Optimal Red Teaming for Large Language Models
What About the Scene With the Hitler Reference? HAUNT: A Framework to Probe LLMs’ Self-consistency in Closed Domains Via Adversarial Nudge
ReasMark: A Robust Watermark for Attributing LLM Reasoning Under Knowledge Distillation Attacks
A Multilingual Social Bias Benchmark Incorporating Thinking Processes
TROJail: Trajectory-Level Optimization for Multi-Turn Large Language Model Jailbreaks with Process Rewards