InjazAI

OpenEvoShield is a co-evolutionary continual defense framework for LLM-based multi-agent systems that outperforms static and continual baselines in detecting unseen attacks while keeping false positive rates low.

1 min read

OpenEvoShield: Dual Non-Stationary Continual Defense for Open-World Multi-Agent System Attacks

Overview

FAQ

What is OpenEvoShield?

A co-evolutionary continual defense framework for LLM-based multi-agent systems to protect against dynamic attacks in open-world environments.

How does OpenEvoShield differ from static defenses?

It adapts to dual drifts in attack strategies and normal agent behavior, while static defenses degrade when distribution shifts.

Can OpenEvoShield be applied in safety-critical applications?

Yes, it is designed for safety-critical applications like multi-agent systems in healthcare or transportation.

Source: arXiv cs.AI

AI-assisted content, human-reviewed.