LLM Safety
LLM: Large Language Model
Overview
LLM safety studies how models can avoid harmful assistance, manipulation, privacy violations, and unreliable advice while still helping with legitimate requests. Approaches include safety training and data filtering, inference-time guardrails and steering, output monitoring, and mechanistic methods that probe or edit internal representations. Benchmarks increasingly test jailbreaks across languages and modalities, long conversations, tool use, and high-stakes settings such as health and cybersecurity. A central challenge is maintaining robustness against adaptive attacks and changing contexts without causing over-refusal or sacrificing utility.
Latest papers
470 papers31 papers in the last four weeks, down 50% on the four weeks before. 0.2% of all new papers.
A Neurosymbolic Approach to Natural Language Formalization and Verification
MENTOR: A Metacognition-Driven Self-Evolution Framework for Uncovering and Mitigating Implicit Domain Risks in LLMs
Moloch's Bargain: Emergent Misalignment When LLMs Compete for Audiences
UpSafeC: Upcycling for Controllable Safety in Large Language Models
Can Large Language Models Develop Gambling Addiction?
Context Misleads LLMs: The Role of Context Filtering in Maintaining Safe Alignment of LLMs
LLMs Encode Harmfulness and Refusal Separately
Response Attack: Exploiting Contextual Priming to Jailbreak Large Language Models
PRISON: Unmasking the Criminal Potential of Large Language Models
HauntAttack: When Attack Follows Reasoning as a Shadow
Popular but Wrong: Understanding and Mitigating LLM Overconfidence through Knowledge Popularity
Decoding One Safety Trigger Token for Balancing Safety and Usability in Large Language Models
LiteLMGuard: Seamless and Lightweight On-Device Guardrails for Small Language Models against Quantization Vulnerabilities
Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models
Divide and Conquer: A Hybrid Strategy Defeats Multimodal Large Language Models
LaSEr-Edit: Localized Span-level Error Editing with Energy-based Localization
Could Mobile AI be Exploited?: Understanding the Exploitations of Trust and Ethics Gap in On-Device Small Language Models
LiSeCo: Linear Semantic Control for Language Generation
Fake News Detectors are Biased against Texts Generated by Large Language Models
GossipCop++ and PolitiFact++, thus amalgamating human-validated articles with LLM-generated fake and real news.