AgentSafeEvo: Adversarial Co-Evolution for LLM Agent Safety

arXiv:2610.19422 · cs.AI · Submitted 2026-10-14 · Read on arXiv

cs.AI

Submitted: 2026-10-14

Updated: 2026-10-14

Code: https://github.com/lalalamdbf/AgentSafeEvohttps:

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Adversarial Co-evolution — 1× in this paper

Sources

• Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks
• Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning
• Hindsight Credit Assignment for Long-Horizon LLM Agents
• Navigating the Risks: A Survey of Security, Privacy, and Ethics Threats in LLM-Based Agents
• Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models
• Safety Alignment of LMs via Non-cooperative Games
• MAGIC: A Co-Evolving Attacker-Defender Adversarial Game for Robust LLM Safety
• Adversarial Reinforcement Learning for Large Language Model Agent Safety
• Purple-teaming LLMs with Adversarial Defender Training
• Agent-SafetyBench: Evaluating the Safety of LLM Agents
• Defending Against Indirect Prompt Injection Attacks With Spotlighting
• Agent Safety Alignment via Reinforcement Learning
• TOUCAN: Synthesizing 1.5M Tool-Agentic Data from Real-World MCP Environments
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• Qwen3 Technical Report
• ALFWorld: Aligning Text and Embodied Environments for Interactive Learning
• AgentRL: Scaling Agentic Reinforcement Learning with a Multi-Turn, Multi-Task Framework
• OpenAI GPT-5 System Card
• LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models
• HybridFlow: A Flexible and Efficient RLHF Framework

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems