Connecting the Dots in Agentic AI Security: A Cross-Dimensional Threat Taxonomy, Evaluation Maturity, and Open Challenges
cs.CR, cs.AI
Submitted: 2026-09-20
Updated: 2026-09-20
Comments: 22 pages
Code: https://github.com/Significant-Gravitas/AutoGPT
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Human Society-Inspired Approaches to Agentic AI Security: The 4C Framework
- A Survey on Agentic Security: Applications, Threats and Defenses
- The Attack and Defense Landscape of Agentic AI: A Comprehensive Survey
- Five Queries Are Enough: Query-Efficient and Surrogate-Free Membership Inference Attacks on RAG via Entailment
- A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration
- Securing Agentic AI: A Comprehensive Threat Model and Mitigation Framework for Generative AI Agents
- MCPXKIT: The Unified Toolkit for Analyzing Model Context Protocol Security
- Ignore Previous Prompt: Attack Techniques For Language Models
- UDora: A Unified Red Teaming Framework against LLM Agents by Dynamically Hijacking Their Own Reasoning
- AgentLAB: Benchmarking LLM Agents against Long-Horizon Attacks
- Demonstrating specification gaming in reasoning models
- Reward Hacking in Language Model Agents: Revisiting AI Safety Gridworlds
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
- Alignment faking in large language models
- Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security
- Overcoming the Retrieval Barrier: Indirect Prompt Injection in the Wild for LLM Systems
- Why Are Web AI Agents More Vulnerable Than Standalone LLMs? A Security Analysis
- Stop Fixating on Prompts: Reasoning Hijacking and Constraint Tightening for Red-Teaming LLM Agents
- ShadowCoT: Cognitive Hijacking for Stealthy Reasoning Backdoors in LLMs
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills
- Token Composition: A Graph Based on EVM Logs