mlsanityData to 7 Oct 2026

Evaluating Generalization Mechanisms in Autonomous Cyber Attack Agents

arXiv 2603.10041 ↗First posted 6 Mar 2026cs.CR cs.LGcited by 0

On the map: LLM agents and tool use → Agent threats, guardrails and governance

Abstract (arXiv)

Autonomous offensive agents often fail to transfer beyond the networks on which they are trained. We isolate a minimal but fundamental shift -- unseen host/subnet IP reassignment in an otherwise fixed enterprise scenario -- and evaluate attacker generalization in the NetSecGame environment. Agents are trained on five IP-range variants and tested on a sixth unseen variant; only the meta-learning agent may adapt at test time. We compare three agent families (traditional RL, adaptation agents, and LLM-based agents) and use action-distribution-based behavioral/XAI analyses to localize failure modes. Some adaptation methods show partial transfer but significant degradation under unseen reassignment, indicating that even address-space changes can break long-horizon attack policies. Under our evaluation protocol and agent-specific assumptions, prompt-driven pretrained LLM agents achieve the highest success on the held-out reassignment, but at the cost of increased inference-time compute, reduced transparency, and practical failure modes such as repetition/invalid-action loops.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use. Builds on 21 of its 41 references that are in the map. First posted on arXiv 6 Mar 2026.

Builds on

21 of 41 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023LLMs Killed the Script Kiddie: How Agents Supported by Large Language Models Change the Landscape of Network Threat Testing10 Oct

Other references · 20

  1. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  2. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  3. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,861
  4. 2013Playing Atari with Deep Reinforcement Learningcited by 3,211
  5. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,175
  6. 2018On First-Order Meta-Learning Algorithmscited by 1,003
  7. 2021A Survey of Zero-shot Generalisation in Deep Reinforcement Learningcited by 164
  8. 2017DARLA: Improving Zero-Shot Transfer in Reinforcement Learningcited by 147
  9. 2023Getting pwn'd by AI: Penetration Testing with Large Language Modelscited by 82
  10. 2022Beyond CAGE: Investigating Generalization of Learned Autonomous Network Defense Policiescited by 22
  11. 2023Zero-Shot Reinforcement Learning from Low Quality Datacited by 21
  12. 2023NASimEmu: Network Attack Simulator & Emulator for Training Agents Generalizing to Novel Scenarioscited by 20
8 more
  1. 2023Out of the Cage: How Stochastic Parrots Win in Cyber Security Environmentscited by 17
  2. 2024Entity-based Reinforcement Learning for Autonomous Cyber Defencecited by 11
  3. 2024Structural Generalization in Autonomous Cyber Incident Response with Message-Passing Neural Networks and Reinforcement Learningcited by 7
  4. 2021Zero-Shot Reinforcement Learning on Graphs for Autonomous Exploration Under Uncertaintycited by 6
  5. 2025Zero-Shot Reinforcement Learning Under Partial Observabilitycited by 6
  6. 2020Zero-Shot Reinforcement Learning with Deep Attention Convolutional Neural Networkscited by 3
  7. 2021Reinforcement Learning for Industrial Control Network Cyber Security Orchestrationcited by 3
  8. 2024Evaluation of Reinforcement Learning for Autonomous Penetration Testing using A3C, Q-learning and DQNcited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Assessing Automated Prompt Injection Attacks in Agentic Environments
  2. 2026In-Context Autonomous Network Incident Response: An End-to-End Large Language Model Agent Approach
  3. 2026Distilling Knowledge from Large Language Models into Lightweight Reinforcement Learning Agents for Autonomous Cyber Operations
  4. 2026To Defend Against Cyber Attacks, We Must Teach AI Agents to Hack
  5. 2025AegisLLM: Scaling Agentic Systems for Self-Reflective Defense in LLM Security
  6. 2026Proteus: A Self-Evolving Red Team for Agent Skill Ecosystems
  7. 2026A Red Teaming Framework for Evaluating Robustness of AI-enabled Security Orchestration, Automation, and Response Systems
  8. 2026Ask the Expert: LLM-Guided Reinforcement Learning for Autonomous Cyber Defense