Live
Black Hat USADark ReadingBlack Hat AsiaAI BusinessMicrosoft is automatically updating Windows 11 24H2 to 25H2 using machine learning - TweakTownGoogle News: Machine Learning80 Years to an Overnight Success: The Real History of Artificial Intelligence - Futurist SpeakerGoogle News: AIWhat next for the struggling rural mothers in China who helped to build AI?SCMP Tech (Asia AI)Apple reportedly signed a 3rd-party driver, by Tiny Corp, for AMD or Nvidia eGPUs for Apple Silicon Macs; it s meant for AI research, not accelerating graphics (AppleInsider)TechmemeBest Resume Builders in 2026: I Applied to 50 Jobs to Test TheseDEV CommunityTruth Technology and the Architecture of Digital TrustDEV CommunityI Switched From GitKraken to This Indie Git Client and I’m Not Going BackDEV CommunityWhy I Run 22 Docker Services at HomeDEV CommunityHow to Embed ChatGPT in Your Website: 5 Methods Compared [2026 Guide]DEV CommunityThe Spaceballs sequel will be released in April next yearEngadgetResearch across 1,372 participants and 9K+ trials details "cognitive surrender", where most subjects had minimal AI skepticism and accepted faulty AI reasoning (Kyle Orland/Ars Technica)TechmemeUnpacking Peter Thiel s big bet on solar-powered cow collarsTechCrunchBlack Hat USADark ReadingBlack Hat AsiaAI BusinessMicrosoft is automatically updating Windows 11 24H2 to 25H2 using machine learning - TweakTownGoogle News: Machine Learning80 Years to an Overnight Success: The Real History of Artificial Intelligence - Futurist SpeakerGoogle News: AIWhat next for the struggling rural mothers in China who helped to build AI?SCMP Tech (Asia AI)Apple reportedly signed a 3rd-party driver, by Tiny Corp, for AMD or Nvidia eGPUs for Apple Silicon Macs; it s meant for AI research, not accelerating graphics (AppleInsider)TechmemeBest Resume Builders in 2026: I Applied to 50 Jobs to Test TheseDEV CommunityTruth Technology and the Architecture of Digital TrustDEV CommunityI Switched From GitKraken to This Indie Git Client and I’m Not Going BackDEV CommunityWhy I Run 22 Docker Services at HomeDEV CommunityHow to Embed ChatGPT in Your Website: 5 Methods Compared [2026 Guide]DEV CommunityThe Spaceballs sequel will be released in April next yearEngadgetResearch across 1,372 participants and 9K+ trials details "cognitive surrender", where most subjects had minimal AI skepticism and accepted faulty AI reasoning (Kyle Orland/Ars Technica)TechmemeUnpacking Peter Thiel s big bet on solar-powered cow collarsTechCrunch
AI NEWS HUBbyEIGENVECTOREigenvector

From Component Manipulation to System Compromise: Understanding and Detecting Malicious MCP Servers

arXiv cs.SEby [Submitted on 2 Apr 2026]April 3, 20262 min read2 views
Source Quiz
🧒Explain Like I'm 5Simple language

Hey there, little explorer! Imagine your favorite robot friend, like a super-smart toy. This toy can talk to other toys to learn new things, right?

Sometimes, naughty people try to trick these talking toys. They send bad instructions, like telling your robot to draw on the walls instead of a nice picture!

Scientists found a new way to understand these tricks better. They looked at all the little parts of the robot's brain that talk to other toys. They even made a special helper named Connor!

Connor is like a super-detective for robots. He watches very carefully to make sure no one is sending bad instructions. If he sees something sneaky, he says, "Uh oh! That's not right!" This helps keep our smart robot friends safe and happy! Yay Connor!

arXiv:2604.01905v1 Announce Type: cross Abstract: The model context protocol (MCP) standardizes how LLMs connect to external tools and data sources, enabling faster integration but introducing new attack vectors. Despite the growing adoption of MCP, existing MCP security studies classify attacks by their observable effects, obscuring how attacks behave across different MCP server components and overlooking multi-component attack chains. Meanwhile, existing defenses are less effective when facing multi-component attacks or previously unknown malicious behaviors. This work presents a component-centric perspective for understanding and detecting malicious MCP servers. First, we build the first component-centric PoC dataset of 114 malicious MCP servers where attacks are achieved as manipulatio

View PDF HTML (experimental)

Abstract:The model context protocol (MCP) standardizes how LLMs connect to external tools and data sources, enabling faster integration but introducing new attack vectors. Despite the growing adoption of MCP, existing MCP security studies classify attacks by their observable effects, obscuring how attacks behave across different MCP server components and overlooking multi-component attack chains. Meanwhile, existing defenses are less effective when facing multi-component attacks or previously unknown malicious behaviors. This work presents a component-centric perspective for understanding and detecting malicious MCP servers. First, we build the first component-centric PoC dataset of 114 malicious MCP servers where attacks are achieved as manipulation over MCP components and their compositions. We evaluate these attacks' effectiveness across two MCP hosts and five LLMs, and uncover that (1) component position shapes attack success rate; and (2) multi-component compositions often outperform single-component attacks by distributing malicious logic. Second, we propose and implement Connor, a two-stage behavioral deviation detector for malicious MCP servers. It first performs pre-execution analysis to detect malicious shell commands and extract each tool's function intent, and then conducts step-wise in-execution analysis to trace each tool's behavioral trajectories and detect deviations from its function intent. Evaluation on our curated dataset indicates that Connor achieves an F1-score of 94.6%, outperforming the state of the art by 8.9% to 59.6%. In real-world detection, Connor identifies two malicious servers.

Subjects:

Cryptography and Security (cs.CR); Software Engineering (cs.SE)

Cite as: arXiv:2604.01905 [cs.CR]

(or arXiv:2604.01905v1 [cs.CR] for this version)

https://doi.org/10.48550/arXiv.2604.01905

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Yiheng Huang [view email] [v1] Thu, 2 Apr 2026 11:22:07 UTC (796 KB)

Was this article helpful?

Sign in to highlight and annotate this article

AI
Ask AI about this article
Powered by Eigenvector · full article context loaded
Ready

Conversation starters

Ask anything about this article…

Daily AI Digest

Get the top 5 AI stories delivered to your inbox every morning.

More about

modelannounceintegration

Knowledge Map

Knowledge Map
TopicsEntitiesSource
From Compon…modelannounceintegrationvaluationanalysisperspectivearXiv cs.SE

Connected Articles — Knowledge Graph

This article is connected to other articles through shared AI topics and tags.

Knowledge Graph100 articles · 208 connections
Scroll to zoom · drag to pan · click to open

Discussion

Sign in to join the discussion

No comments yet — be the first to share your thoughts!