Detecting and Mitigating Harmful Content
Research on identifying and preventing harmful or malicious content generated by or input to LLMs
Detecting and Mitigating Harmful Content
Research on Large Language Models in Detecting and Mitigating Harmful Content
Combating Hate Speech with Multi-Task Learning
Improving generalization for detecting harmful content targeting political figures
Fighting Bias in AI Language Models
A toolkit to detect and mitigate prediction biases in LLMs
Profiling Hate Speech Spreaders Online
Understanding who reshares hate content on social media
Meme Safety for AI Systems
Evaluating how multimodal models respond to harmful meme content
ChatGPT vs. Spam: New Frontiers in Email Security
Leveraging LLMs to enhance spam detection accuracy
Securing AI Art Generation
A robust approach to filtering harmful concepts in text-to-image models
Combat Fake News with LLMs
Using Defense Among Competing Wisdom framework to explain detection results
ProFS: A Better Way to Reduce LLM Toxicity
Making LLMs safer through targeted model editing
S-Eval: Securing the Future of LLMs
A systematic framework for comprehensive LLM safety evaluation
AI Threats to Election Integrity
Mapping GenAI's dangerous potential in electoral manipulation
The Evolution of Deepfake Detection
From Single-Modal to Multi-Modal Approaches for Enhanced Security
Detecting Mixed Reality Fakes
First benchmark for multimodal misinformation from multiple sources
Fact-Checking Our AI Visionaries
A new benchmark for evaluating factuality in multimodal AI systems
Combating Fake News with AI Intelligence
A Knowledge-guided Framework for Few-shot Detection
Harnessing LVLMs for Fake News Detection
How Visual-Language Models Outperform in Multimodal Misinformation Classification
Smart Danmaku Moderation for Video Platforms
Using LLMs and Impact Captions to Reduce Toxic Comments
Combating Fake News with Adaptive AI
Using dynamic analysis to detect misinformation across platforms
Bypassing LLM Safety Filters
Critical vulnerability found in safety evaluation metrics
Measuring LLM Safety Through Offensive Content Progression
A new approach to benchmarking model sensitivity to harmful content
Human-Guided Data Exploration
Collaborative AI Safety Through Interactive Data Augmentation
Lightweight Security for LLMs
Making AI safety models smaller without sacrificing performance
Proactive LLM Safety Auditing
A novel approach for detecting catastrophic AI responses before they cause harm
Reliable Guardrails for LLMs
Improving calibration of content moderation systems
Small Models, Big Security Impact
How fine-tuned SLMs outperform larger models in content moderation
Beyond Binary: Fine-Grained LLM Content Detection
Recognizing the spectrum of human-AI collaboration in text
Fighting Misinformation with Smarter Detection
A Knowledge-Based Approach Using Annotator Reliability
Detecting Out-of-Scope Questions in LLMs
New resources to prevent AI hallucinations when questions seem relevant
Fooling LLM Detectors: The Proxy Attack Strategy
How simple attacks can make AI-generated text pass as human-written
AI-Generated Fake News in the LLM Era
Assessing human and AI detection capabilities against LLM-crafted misinformation
Real-Time LLM Detection
A New Betting-Based Approach to Identify AI-Generated Content as it Arrives
The Reality Gap in LLM Text Detection
Why current detectors fail in real-world scenarios
Crossing Cultural Boundaries in Hate Speech Detection
First multimodal, multilingual hate speech dataset with multicultural annotations
Combating Online Hate with AI
Leveraging GPT-3.5 Turbo to detect and mitigate hate speech on X (Twitter)
Smarter Toxicity Detection in Memes
Using Knowledge Distillation and Infusion to Combat Online Toxicity
Developing Smarter LLM Guardrails
A flexible methodology for detecting off-topic prompts without real-world data
Fighting Rumors with Network Science
A Novel Epidemic-Inspired Approach to Rumor Detection
Smarter Security Testing for AI Image Generators
Using LLMs to systematically find vulnerabilities in text-to-image models
Securing LLMs Against Harmful Content
A Dynamic Filtering Approach Without Retraining
Uncovering Hidden Messages Online
Novel methodology for detecting coded 'dog whistles' in digital content
Combating Evolving Toxic Content
Adaptable Detection Systems for Security in LLMs
Combating Organized Disinformation
Network-informed prompt engineering for detecting astroturfing in social media
Unmasking Deceptive UI Designs
Automated Detection of Manipulation in User Interfaces
Harnessing LLMs to Combat Climate Misinformation
Using AI with human oversight to detect false climate claims
HateBench: Evaluating Hate Speech Detection Against LLM Threats
First comprehensive benchmark for testing detectors against LLM-generated hate content
Modeling Self-Destructive Reasoning in LLMs
A mathematical framework for tracking toxicity amplification in language models
Emotion-Aware Cyberbullying Detection
Beyond Binary Classification: Detecting Harassment and Defamation
Securing AI: Advanced Safety Testing for LLMs
Automating comprehensive safety evaluation with ASTRAL
The Dark Side of Persuasion
How LLMs Use Personalization and False Statistics to Change Minds
Automating Counterspeech Evaluation
A novel framework for measuring effectiveness in combating hate speech
Safeguarding AI: Pre-Deployment Testing of LLMs
Insights from External Safety Evaluation of OpenAI's o3-mini Model
Divergent Emotional Patterns in Disinformation on Social Med...
By Iván Arcos, Paolo Rosso...
Detecting LLM-Laundered Fake News
How AI-paraphrased misinformation evades current detection systems
Challenges and Innovations in LLM-Powered Fake News Detectio...
By Jingyuan Yi, Zeqiu Xu...
SafeSwitch: Smarter AI Safety Controls
Using internal activation patterns to regulate LLM behavior without sacrificing capability
Almost Surely Safe Alignment of Large Language Models at Inf...
By Xiaotong Ji, Shyam Sundhar Ramesh...
Evaluating the Blind Spots of LLM Safety
Can larger models accurately detect harm in smaller ones?
Media Bias Detector: Designing and Implementing a Tool for R...
By Jenny S Wang, Samar Haider...
LLMs and Annotation Disagreement
How AI handles ambiguity in offensive language detection
Adaptive Prompting: Ad-hoc Prompt Composition for Social Bia...
By Maximilian Spliethöver, Tim Knebler...
Balancing AI Safety and Scientific Freedom
A benchmark for evaluating LLM safety mechanisms against dual-use risks
Intent-Aware Repair for Safer LLMs
Precision-targeting toxic behaviors without compromising model capabilities
Securing LLM Interactions: The Guardrail Approach
A comprehensive safety pipeline for trustworthy AI interactions
Beyond Binary: Tackling Hate Speech Detection Challenges
Innovative approaches to handle annotator disagreement in content moderation
FLAME: Flexible LLM-Assisted Moderation Engine
By Ivan Bakulin, Ilia Kopanichuk...
Demystifying Hateful Content: Leveraging Large Multimodal Mo...
By Ming Shan Hee, Roy Ka-Wei Lee
VLDBench: Vision Language Models Disinformation Detection Be...
By Shaina Raza, Ashmal Vayani...
Unlocking LLM Transparency with Sparse Autoencoders
Optimizing interpretable features for critical classifications
Combating Misinformation with AI
Introducing HintsOfTruth: A Multimodal Dataset for Checkworthiness Detection
Securing LLMs Against Unsafe Prompts
A Novel Gradient Analysis Approach with Minimal Reference Data
AI-Powered Deception Detection in Negotiations
Using Game Theory to Unmask 'Too Good to Be True' Offers
Smart Safety Guardrails for LLMs
Optimizing security without compromising performance
Combating Hateful Memes with Advanced AI
Breakthrough in fine-tuning multimodal models for online safety
ThinkGuard: Deliberative Safety for LLMs
Enhancing AI guardrails through slow, deliberative thinking
Advancing Extreme Speech Detection
Comparing Open-Source vs. Proprietary LLMs for Content Moderation
Red Flag Tokens: A New Approach to LLM Safety
Enhancing harmfulness detection without compromising model capabilities
Explainable Propaganda Detection
Using LLMs to detect and explain propagandistic content
MemeIntel: Smart Detection of Harmful Memes
AI-powered explainable detection system for propaganda and hate speech
Detecting Disguised Toxic Content with AI
Using LLMs to Extract Effective Search Queries
Geographically-Aware Hate Speech Detection
Evaluating LLMs for culturally contextualized content moderation
Controlling Toxic AI Outputs
A statistical approach to safer large language models
Balancing Safety and Utility in AI Role-Playing
New frameworks for managing dangerous content in character simulations
Safety Risks in LLM Role-Play
Uncovering & addressing security vulnerabilities when LLMs assume character roles
Fighting Fake News with AI
Comparing LLM-based strategies for misinformation detection
Uncovering Hidden Toxicity in Language
A novel approach to detecting implicit harmful content in LLMs
Filtering Harm in LLM Training Data
Evaluating safety strategies and their implications for vulnerable groups
SafeSpeech: Detecting Toxicity Across Conversations
Beyond message-level analysis to context-aware toxic language detection
Smarter Content Moderation for LLMs
Risk-level assessment for safer AI platforms
AI-Powered Violence Detection in Historical Texts
Large Language Models Automate Analysis of Ancient Violence
Evaluating AI Models for Inclusive Computing Language
Benchmarking LLMs' ability to detect harmful technical terminology
Protecting Children in the Age of AI
New benchmark for evaluating LLM content risks for minors
Simulating Moderation at Scale
Using LLMs to evaluate online content moderation strategies
Securing Code LLMs Against Harmful Content
A novel automated framework for robust content moderation in code generation
Fixing False Refusals in AI Safety
Making LLMs smarter about when to say 'no'
Detecting Coded Islamophobia Online
Using LLMs to Identify and Analyze Extremist Language
Harnessing LLMs for Bug Report Analysis
Using AI to extract failure-inducing inputs from natural language bug reports
Breaking Language Barriers in Content Moderation
Adapting LLMs for Low-Resource Languages: The Sinhala Case Study
Enhanced Harmful Content Detection
Combining LLMs with Knowledge Graphs for Safer AI Systems
Multi-label Hate Speech Detection
Advancing beyond binary classification for more effective content moderation
Protecting AI Across Languages
A Multilingual Approach to LLM Content Moderation
Defending Against LLM Jailbreaks
A Feature-Aware Approach to Detecting and Mitigating Harmful Outputs
Detecting Sexism Across Modalities
First multimodal Spanish dataset for sexism detection in social media videos
