The field of Large Language Models (LLMs) is rapidly evolving, with a growing focus on safety and security. Recent research has highlighted the vulnerabilities of LLMs to various types of attacks, including jailbreak attacks, cross-user poisoning, and adversarial attacks. These attacks can be used to bypass safety filters, generate harmful content, and exploit user interactions. To address these challenges, researchers are developing new defense strategies, such as prompt-level defenses, model-level defenses, and training-time interventions. Noteworthy papers in this area include 'Evaluating Adversarial Vulnerabilities in Modern Large Language Models' and 'Defending Large Language Models Against Jailbreak Exploits with Responsible AI Considerations', which demonstrate the effectiveness of these defense strategies in mitigating the risks associated with LLMs. Overall, the field is moving towards a more robust and secure deployment of LLMs, with a focus on responsible AI development and deployment.
Advances in Large Language Model Safety and Security
Sources
Evaluating Adversarial Vulnerabilities in Modern Large Language Models
MURMUR: Using cross-user chatter to break collaborative language agents in groups
Shadows in the Code: Exploring the Risks and Defenses of LLM-based Multi-Agent Software Development Systems
TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization
RoguePrompt: Dual-Layer Ciphering for Self-Reconstruction to Circumvent LLM Moderation
Defending Large Language Models Against Jailbreak Exploits with Responsible AI Considerations
Understanding and Mitigating Over-refusal for Large Language Models via Safety Representation
Can LLMs Threaten Human Survival? Benchmarking Potential Existential Threats from LLMs via Prefix Completion
Adversarial Attack-Defense Co-Evolution for LLM Safety Alignment via Tree-Group Dual-Aware Search and Optimization
Building Resilient Information Ecosystems: Large LLM-Generated Dataset of Persuasion Attacks
Automating Deception: Scalable Multi-Turn LLM Jailbreaks
An Invariant Latent Space Perspective on Language Model Inversion
Prompt Fencing: A Cryptographic Approach to Establishing Security Boundaries in Large Language Model Prompts
Cross-LLM Generalization of Behavioral Backdoor Detection in AI Agent Supply Chains
PropensityBench: Evaluating Latent Safety Risks in Large Language Models via an Agentic Approach
Large Language Models' Complicit Responses to Illicit Instructions across Socio-Legal Contexts
Semantic Superiority vs. Forensic Efficiency: A Comparative Analysis of Deep Learning and Psycholinguistics for Business Email Compromise Detection
Self-Guided Defense: Adaptive Safety Alignment for Reasoning Models via Synthesized Guidelines
Constructing and Benchmarking: a Labeled Email Dataset for Text-Based Phishing and Spam Detection Framework