Review maps 284 attacks on agentic AI systems
Original title: Agents of Chaos: A Systematization of Knowledge (SoK) on Malicious Attacks on Agentic AI Systems
Tool, MCP, and action-layer attacks dominate, while only 278 of 660 attack-safeguard links are directly assessed.
A systematization of knowledge reviews malicious attacks on agentic AI systems that combine language models with tools, memory, retrieval, code execution, and multi-agent coordination. The authors screened 186 sources, retained 98, and coded 284 atomic attacks, 452 atomic safeguards, and 660 attack-safeguard relationships. Tool, MCP, and action-layer attacks form the largest share, while prompt injection and action hijacking have the broadest paper coverage. Safeguard coverage is uneven, with strong coverage uncommon.
- The review retained 98 of 186 screened candidate sources.
- Authors identified 452 atomic safeguards and 660 explicit relationships.
- Only 278 of 660 attack-safeguard relationships are directly assessed.
- Attacks are organized into eight causal-mechanism classes.
From the original article
1 min read at arxiv.orgAgentic AI systems extend large language models with tools, memory, retrieval, browser or computer-use interfaces, code execution, external services, and multi-agent coordination.
By Thanh M. Truong-Le · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.