The field of large language models (LLMs) is rapidly evolving, with a growing focus on security, trustworthiness, and reliability. Recent research has highlighted the importance of developing LLMs that can operate in a secure and transparent manner, particularly in high-stakes applications such as finance, healthcare, and industrial automation. One of the key challenges in this area is the need to balance the benefits of LLMs, such as their ability to process and generate human-like language, with the risks associated with their use, such as the potential for bias, misuse, and exploitation. To address these challenges, researchers are exploring a range of innovative solutions, including the development of novel architectures, such as graph-based and attention-based models, and the application of advanced techniques, such as reinforcement learning and mechanistic interpretability. Noteworthy papers in this area include 'ADA: Automated Moving Target Defense for AI Workloads via Ephemeral Infrastructure-Native Rotation in Kubernetes', which introduces a novel approach to securing AI workloads using automated moving target defense, and 'SentinelAgent: Graph-based Anomaly Detection in Multi-Agent Systems', which presents a system-level anomaly detection framework for multi-agent systems.
Advances in Secure and Trustworthy Large Language Models
Sources
ADA: Automated Moving Target Defense for AI Workloads via Ephemeral Infrastructure-Native Rotation in Kubernetes
System Prompt Extraction Attacks and Defenses in Large Language Models
Scalable, Symbiotic, AI and Non-AI Agent Based Parallel Discrete Event Simulations
Seven Security Challenges That Must be Solved in Cross-domain Multi-agent LLM Systems
LLM Agents Should Employ Security Principles
SentinelAgent: Graph-based Anomaly Detection in Multi-Agent Systems
An Adversary-Resistant Multi-Agent LLM System via Credibility Scoring
Beyond the Black Box: Interpretability of LLMs in Finance
So, I climbed to the top of the pyramid of pain -- now what?
STRATUS: A Multi-agent System for Autonomous Reliability Engineering of Modern Clouds
Towards Secure MLOps: Surveying Attacks, Mitigation Strategies, and Research Challenges
Beyond the Protocol: Unveiling Attack Vectors in the Model Context Protocol Ecosystem
Docker under Siege: Securing Containers in the Modern Era
Composable Building Blocks for Controllable and Transparent Interactive AI Systems
Attention Knows Whom to Trust: Attention-based Trust Management for LLM Multi-Agent Systems
ATAG: AI-Agent Application Threat Assessment with Attack Graphs
Feedstack: Layering Structured Representations over Unstructured Feedback to Scaffold Human AI Conversation
Sampling Preferences Yields Simple Trustworthiness Scores
Robustness of Prompting: Enhancing Robustness of Large Language Models Against Prompting Attacks
From Theory to Practice: Real-World Use Cases on Trustworthy LLM-Driven Process Modeling, Prediction and Automation
Privacy and Security Threat for OpenAI GPTs
TRiSM for Agentic AI: A Review of Trust, Risk, and Security Management in LLM-based Agentic Multi-Agent Systems
Contextual Integrity in LLMs via Reasoning and Reinforcement Learning
HADA: Human-AI Agent Decision Alignment Architecture
Demonstrations of Integrity Attacks in Multi-Agent Systems
Agentic AI for Intent-Based Industrial Automation
SECNEURON: Reliable and Flexible Abuse Control in Local LLMs via Hybrid Neuron Encryption
Control Tax: The Price of Keeping AI in Check