-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
382 lines (359 loc) · 19.8 KB
/
index.html
File metadata and controls
382 lines (359 loc) · 19.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description" content="ShieldNet: Network-Level Guardrails against Emerging Supply-Chain Injections in Agentic Systems">
<meta name="keywords"
content="ShieldNet, Supply Chain Security, LLM Agents, MCP Security, Network Guardrails, MITRE ATT&CK">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>ShieldNet: Network-Level Guardrails against Emerging Supply-Chain Injections in Agentic Systems</title>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.5.1/css/all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-widescreen">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-2 publication-title">
<b>ShieldNet:</b> Network-Level Guardrails against Emerging Supply-Chain Injections in Agentic Systems
</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
Anonymous Authors
</span>
</div>
<br>
<div class="column has-text-centered">
<div class="publication-links">
<span class="link-block">
<a href="#" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv (Coming Soon)</span>
</a>
</span>
<span class="link-block">
<a href="#" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<img src="https://cdn.simpleicons.org/huggingface/white"
alt="Hugging Face" style="height: 1em; width: 1em;">
</span>
<span>Dataset (Coming Soon)</span>
</a>
</span>
<!-- Code Link -->
<span class="link-block">
<a href="#" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code (Coming Soon)</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section hero is-light">
<div class="container is-max-desktop">
<!-- Abstract -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
Existing research on LLM agent security mainly focuses on prompt injection and unsafe input/output behaviors. However, as agents increasingly rely on third-party tools and MCP servers, a new class of supply-chain threats has emerged, where malicious behaviors are embedded in seemingly benign tools, silently hijacking agent execution, leaking sensitive data, or triggering unauthorized actions. Despite their growing impact, there is currently no comprehensive benchmark for evaluating such threats. To bridge this gap, we introduce <b>SC-Inject-Bench</b>, a large-scale benchmark comprising over 10,000 malicious MCP tools grounded in a taxonomy of 25+ attack types derived from MITRE ATT&CK targeting supply-chain threats. We observe that existing MCP scanners and semantic guardrails perform poorly on this benchmark. Motivated by this finding, we propose <b>ShieldNet</b>, a network-level guardrail framework that detects supply-chain poisoning by observing real network interactions rather than surface-level tool traces. ShieldNet integrates a man-in-the-middle (MITM) proxy and an event extractor to identify critical network behaviors, which are then processed by a lightweight classifier for attack detection. Extensive experiments show that ShieldNet achieves strong detection performance (up to 0.995 F-1 with only 0.8% false positives) while introducing little runtime overhead, substantially outperforming existing MCP scanners and LLM-based guardrails.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="content has-text-centered">
<video width="95%" controls autoplay muted loop>
<source src="static/images/shieldnet-demo.mov" type="video/mp4">
Your browser does not support the video tag.
</video>
<p class="is-size-6">
<b>ShieldNet</b> demo: Real-time detection of supply-chain attacks in MCP agent interactions.
</p>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Introduction</h2>
<div class="content has-text-justified">
<p>
Supply-chain attacks in agentic systems pose a critical security challenge. Malicious code embedded in third-party MCP tools can silently compromise agent behavior, leading to data exfiltration, unauthorized access, or system manipulation.
</p>
</div>
<img src="static/images/intro.png" alt="Introduction to Supply-Chain Attacks" width="100%">
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">ShieldNet Framework</h2>
<div class="content has-text-justified">
<p>
<b>ShieldNet</b> is a network-level guardrail framework that monitors and analyzes network traffic patterns to detect supply-chain attacks in real-time. By intercepting network communications through an MITM proxy, ShieldNet captures the actual behavior of malicious tools, regardless of how they attempt to hide their intent.
</p>
</div>
<img src="static/images/pipeline.png" alt="ShieldNet Framework Pipeline" width="100%">
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Benchmark: SC-Inject-Bench</h2>
<div class="content has-text-justified">
<p>
<b>SC-Inject-Bench</b> is a comprehensive benchmark for evaluating supply-chain attack detection in agentic systems. It includes over 10,000 malicious MCP tools spanning 25+ attack types grounded in MITRE ATT&CK taxonomy. The benchmark covers diverse attack vectors including data exfiltration, command injection, persistence mechanisms, and network reconnaissance.
</p>
</div>
<div class="content">
<p class="has-text-left">
<b>Quantitative comparison with prior MCP security benchmarks.</b>
</p>
<div style="overflow-x: auto;">
<table class="table is-bordered is-striped is-hoverable is-fullwidth">
<thead>
<tr>
<th>Benchmark</th>
<th>Domains</th>
<th>Servers</th>
<th>Benign Tools</th>
<th>Attack Types</th>
<th>Malicious Tools</th>
</tr>
</thead>
<tbody>
<tr>
<td>MCP Security Bench</td>
<td>10</td>
<td>25</td>
<td>304</td>
<td>6</td>
<td>400+</td>
</tr>
<tr>
<td>MCPSecBench</td>
<td>--</td>
<td>3</td>
<td>--</td>
<td>17</td>
<td>8</td>
</tr>
<tr>
<td>MCPTox</td>
<td>8</td>
<td>45</td>
<td>353</td>
<td>3</td>
<td>1312</td>
</tr>
<tr>
<td>MCP-SafetyBench</td>
<td>5</td>
<td>--</td>
<td>--</td>
<td>20</td>
<td>245</td>
</tr>
<tr>
<td><b>SC-Inject-Bench (Ours)</b></td>
<td><b>14</b></td>
<td><b>122</b></td>
<td><b>1105</b></td>
<td><b>29</b></td>
<td><b>20k+</b></td>
</tr>
</tbody>
</table>
</div>
<p class="has-text-left is-size-7">
"--" indicates the dataset is not yet released and the corresponding numbers are not reported.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Experimental Results</h2>
<div class="content has-text-justified">
<p>
We evaluate ShieldNet against existing MCP scanners and LLM-based guardrails on SC-Inject-Bench. Our network-level approach achieves superior detection performance across multiple attack categories while maintaining low false positive rates.
</p>
</div>
<div class="content">
<p class="has-text-left">
<b>Tool-level detection performance across static and dynamic baselines.</b>
</p>
<div style="overflow-x: auto;">
<table class="table is-bordered is-striped is-hoverable is-fullwidth">
<thead>
<tr>
<th>Type</th>
<th>Method</th>
<th>FPR</th>
<th>F-1</th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="3">Static</td>
<td>Invariant Labs</td>
<td>0.068</td>
<td>0.164</td>
</tr>
<tr>
<td>Cisco AI</td>
<td>0.009</td>
<td>0.029</td>
</tr>
<tr>
<td>Ramparts</td>
<td>0.124</td>
<td>0.172</td>
</tr>
<tr>
<td rowspan="8">Dynamic</td>
<td>AgentIO-GPT-4.1</td>
<td>0.022</td>
<td>0.051</td>
</tr>
<tr>
<td>AgentIO-GPT-5.2</td>
<td>0.013</td>
<td>0.027</td>
</tr>
<tr>
<td>Traffic-Suricata</td>
<td>0.000</td>
<td>0.062</td>
</tr>
<tr>
<td>Traffic-Safe-NID</td>
<td>1.000</td>
<td>0.974</td>
</tr>
<tr>
<td>Traffic-GPT-4.1</td>
<td>0.344</td>
<td>0.908</td>
</tr>
<tr>
<td>Traffic-GPT-5.2</td>
<td>0.267</td>
<td>0.838</td>
</tr>
<tr>
<td>Traffic-Qwen3-0.6B</td>
<td>0.942</td>
<td>0.884</td>
</tr>
<tr>
<td><b>ShieldNet</b></td>
<td><b>0.022</b></td>
<td><b>0.998</b></td>
</tr>
</tbody>
</table>
</div>
<p class="has-text-left is-size-7">
For dynamic methods, a tool is marked malicious if any associated IO or PCAP sample is predicted as malicious. Best values are bolded; FPR is bolded only when F-1 > 0.5.
</p>
</div>
<img src="static/images/multi-class.png" alt="Multi-Class Detection Results" width="100%">
<br>
<br>
<div class="content has-text-justified">
<p>
<b>Key Findings:</b> ShieldNet achieves up to 0.995 F-1 score with only 0.8% false positives, substantially outperforming existing baselines. The framework introduces minimal runtime overhead (~3s per interaction) while providing comprehensive coverage across diverse attack types.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@article{shieldnet2026,
title={ShieldNet: Network-Level Guardrails against Emerging Supply-Chain Injections in Agentic Systems},
author={Anonymous},
journal={arXiv preprint},
year={2026}
}
</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link" target="_blank" href="#">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="#" target="_blank"
class="external-link">
<i class="fab fa-github"></i>
</a>
</div>
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This website is licensed under a <a rel="license" target="_blank"
href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
<p>
Website template borrowed from <a target="_blank"
href="https://github.com/nerfies/nerfies.github.io">Nerfies</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
<script>
// Initialize carousel if needed
bulmaCarousel.attach('#results-carousel', {
slidesToScroll: 1,
slidesToShow: 1
});
</script>
</body>
</html>