Agents of Chaos: A Systematization of Knowledge (SoK) on Malicious Attacks on Agentic AI Systems
cs.CR
Submitted: 2026-11-07
Updated: 2026-11-07
Terminology
Sources
- SoK: The Attack Surface of Agentic AI - Tools and Autonomy
- The Attack and Defense Landscape of Agentic AI: A Comprehensive Survey
- Prompt Injection attack against LLM-integrated Applications
- Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
- Cloak and Detonate: Scanner Evasion and Dynamic Detection of Agent Skill Malware
- Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems
- Bypassing AI Control Protocols via Agent-as-a-Proxy Attacks
- Importing Phantoms: Measuring LLM Package Hallucination Vulnerabilities
- When the Manual Lies: A Realistic Benchmark to Evaluate MCP Poisoning Attacks for LLM Agents
- Universal and Transferable Adversarial Attacks on Aligned Language Models
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models
- ToolTweak: An Attack on Tool Selection in LLM-based Agents
- A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework
- Security of AI Agents
- MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval
- MemLineage: Lineage-Guided Enforcement for LLM Agent Memory
- Retrieval Observability Bounds on Provenance Detection for Agent Memory Poisoning: Measured Coverage and a Falsified Standalone Detector
- Semantic Attacks on Tool-Augmented LLMs: Securing the Model Context Protocol Against Descriptor-Level Manipulation
- Skills Are Not Islands: Measuring Dependency and Risk in Agent Skill Supply Chains
- Agentic AI for Autonomous Defense in Software Supply Chain Security: Beyond Provenance to Vulnerability Mitigation
Related papers
- From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment
- Privacy Leakage from Recurrent State in LLM Inference
- Incrementally Verifiable Oracle Aided Computation
- Privilege and confidentiality in generative AI workflows
- BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts
- CoVault: Secure, Scalable Analytics of Personal Data