-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
683 lines (620 loc) · 32.7 KB
/
index.html
File metadata and controls
683 lines (620 loc) · 32.7 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
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="google-site-verification" content="pmVSIfR7AGw-ASjuWs2iq3obODI8Iwi7UMugsFenY2Y" />
<!-- SEO Meta Tags -->
<title>Changyi Li (李长艺) | Fudan University | AI Safety & Alignment</title>
<!-- Favicon -->
<link rel="icon" type="image/x-icon" href="favicon.ico?v=20260317-7">
<link rel="shortcut icon" href="favicon.ico?v=20260317-7">
<link rel="icon" type="image/svg+xml" href="favicon.svg?v=20260317-7">
<meta name="description"
content="Changyi Li (李长艺) is a PhD student at Fudan University researching AI safety, alignment, and trustworthy AI systems, advised by Xudong Pan and Min Yang.">
<meta name="keywords"
content="Changyi Li, 李长艺, Fudan University, AI Safety, AI Alignment, LLM Safety, ReasoningShield, AutoControl Arena, Machine Learning, Deep Learning, 复旦大学">
<meta name="author" content="Changyi Li">
<meta name="robots" content="index, follow">
<!-- Open Graph / Social Media Meta Tags -->
<meta property="og:type" content="website">
<meta property="og:url" content="https://cosmosyi.github.io/">
<meta property="og:title" content="Changyi Li (李长艺) | Fudan University | AI Safety & Alignment">
<meta property="og:description"
content="PhD student researching AI Safety and Alignment at Fudan University. Publications on ReasoningShield, AutoControl Arena, and AI Self-Replication.">
<meta property="og:site_name" content="Changyi Li's Homepage">
<!-- Twitter Card Meta Tags -->
<meta name="twitter:card" content="summary">
<meta name="twitter:title" content="Changyi Li (李长艺) | Fudan University | AI Safety & Alignment">
<meta name="twitter:description" content="PhD student researching AI Safety and Alignment at Fudan University.">
<!-- Canonical URL -->
<link rel="canonical" href="https://cosmosyi.github.io/">
<!-- Structured Data (JSON-LD) for Google -->
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "WebSite",
"url": "https://cosmosyi.github.io/",
"name": "Changyi Li's Homepage"
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "ProfilePage",
"url": "https://cosmosyi.github.io/",
"dateModified": "2026-04-03T00:00:00+08:00",
"mainEntity": {
"@id": "https://cosmosyi.github.io/#person",
"@type": "Person",
"name": "Changyi Li",
"alternateName": "李长艺",
"url": "https://cosmosyi.github.io/",
"mainEntityOfPage": "https://cosmosyi.github.io/",
"description": "PhD student at Fudan University researching AI safety, alignment, and trustworthy AI systems.",
"image": "https://cosmosyi.github.io/images/changyi.png",
"jobTitle": "PhD Student",
"affiliation": {
"@type": "Organization",
"name": "Fudan University",
"url": "https://www.fudan.edu.cn/"
},
"alumniOf": {
"@type": "Organization",
"name": "Qilu University of Technology"
},
"email": "24212010017@m.fudan.edu.cn",
"sameAs": [
"https://scholar.google.com/citations?user=AEIJU60AAAAJ",
"https://github.com/CosmosYi"
],
"knowsAbout": [
"AI Safety",
"AI Alignment",
"Trustworthy AI",
"Large Language Models",
"AI Agent",
"Deep Learning",
"Machine Learning"
]
}
}
</script>
<meta name="generator" content="HTML Tidy for Linux/x86 (vers 11 2007), see www.w3.org">
<style type="text/css">
/* Font: Advanced modern UI font - Inter */
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap');
body,
td,
th,
p {
font-family: 'Inter', -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif;
font-size: 16px;
font-weight: 400;
color: #000;
line-height: 1.6;
text-align: justify;
}
a {
color: #1772d0;
text-decoration: none;
transition: color 0.2s;
}
a:focus,
a:hover {
color: #f09228;
text-decoration: underline;
}
/* Headings */
heading {
font-family: 'Inter', -apple-system, sans-serif;
font-size: 18px;
font-weight: 700;
color: #1772d0;
}
h2.sectionheading {
font-family: 'Inter', -apple-system, sans-serif;
font-size: 24px;
font-weight: 700;
color: #222;
border-bottom: 2px solid #eee;
padding-bottom: 5px;
display: inline-block;
margin-bottom: 10px;
}
h1.pageheading {
font-family: 'Playfair Display', serif;
font-size: 38px;
font-weight: 700;
color: #000000;
letter-spacing: -0.5px;
}
hr {
border: 0;
height: 1px;
background-image: none;
background-color: #ddd;
/* Subtle grey lines */
margin: 20px 0;
}
strong,
b {
font-weight: 700;
color: #111;
}
strongred {
color: #1772d0;
font-weight: 700;
}
.image-container img {
border-radius: 15px;
box-shadow: 0 4px 12px rgba(0, 0, 0, 0.1);
}
/* News Items */
.news-container {
margin-top: 20px;
}
.news-item {
padding: 8px 12px;
margin-bottom: 8px;
background-color: #f8faff;
border-left: 3px solid #1772d0;
border-radius: 0 4px 4px 0;
display: flex;
align-items: baseline;
}
.news-date {
color: #1772d0;
font-weight: 700;
min-width: 90px;
margin-right: 15px;
font-size: 0.9em;
font-family: 'Roboto Mono', monospace;
}
.news-content {
flex: 1;
}
/* Buttons */
.pub-buttons {
margin-top: 8px;
}
.pub-button {
display: inline-block;
padding: 4px 12px;
margin-right: 8px;
border-radius: 20px;
/* Pill shaped buttons */
font-size: 0.85em;
font-weight: 500;
border: 1px solid #1772d0;
}
.pub-button.primary {
background-color: #1772d0;
color: white;
}
.pub-button.primary:hover {
background-color: #f09228;
border-color: #f09228;
text-decoration: none;
}
.pub-button.text {
background-color: transparent;
color: #1772d0;
border: 1px solid transparent;
}
.pub-button.text:hover {
background-color: #fdf5eb;
text-decoration: none;
}
/* Paper Title */
.paper-title {
font-size: 1.1em;
color: #1772d0;
font-weight: 700;
}
/* Abstract Box */
.paper pre {
background-color: #f4f4f4;
padding: 10px;
border-radius: 5px;
border: 1px solid #ddd;
font-size: 0.9em;
overflow-x: auto;
}
/* Mobile Responsiveness */
@media (max-width: 800px) {
body {
padding: 0 15px;
}
.main-container {
width: 100% !important;
max-width: 900px !important;
}
/* Stack table layout vertically */
table,
tbody,
tr,
td {
display: block;
width: 100% !important;
box-sizing: border-box;
}
/* Reduce padding on mobile */
td[width="25%"],
td[width="75%"] {
padding: 10px 0 !important;
}
/* Center images in bio and publications sections */
.image-container img,
td[width="25%"] img {
margin: 0 auto 15px auto !important;
display: block;
max-width: 100%;
height: auto;
}
table[id="about"] {
margin-top: 20px !important;
}
/* Refine spacing for elements */
h1.pageheading {
font-size: 30px;
}
h2.sectionheading {
font-size: 22px;
}
}
</style>
<link
href="https://fonts.googleapis.com/css2?family=Roboto+Mono:wght@400;500&family=Playfair+Display:wght@400;700&family=Manrope:wght@600;700;800&display=swap"
rel="stylesheet">
</head>
<body>
<table class="main-container" width="100%" style="max-width: 900px;" border="0" align="center" cellspacing="0"
cellpadding="0"> <!-- Narrower width for cleaner look -->
<tr>
<td>
<!-- Bio Section -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20" style="margin-top: 40px;"
id="about">
<tr>
<td width="75%" valign="middle">
<h1 class="pageheading" style="margin: 0;">Changyi Li (李长艺)</h1>
<p style="font-size: 1.3em; color: #161616; font-weight: 700; margin-top: 10px; margin-bottom: 15px;">
PhD Student at Fudan University
</p>
<p>
I am a incoming PhD student at the <a href="https://secsys.fudan.edu.cn/">System Software and
Security Lab</a>, <a href="https://cs.fudan.edu.cn/">School of Computer Science</a>, <a
href="https://www.fudan.edu.cn/">Fudan University</a>, advised by <a
href="https://ravensanstete.github.io/en/">Xudong Pan</a> and <a
href="https://min-yang-fudan.github.io/">Min Yang</a>.
</p>
<p>
My research focuses on building capable, safe and trustworthy AI systems. Broadly, I explore two
complementary directions: (1) developing learning and optimization methods that strengthen model
competence and generalization, and (2) designing evaluation and alignment approaches that make advanced
AI systems safer, more controllable, and better grounded in human values and knowledge.
</p>
<p style="font-size: 1em; color: #000; margin-top: 15px; margin-bottom: 8px;">
<strong>Email:</strong> 24212010017 [AT] m.fudan.edu.cn
</p>
</td>
<td width="25%" align="center" valign="bottom" class="image-container" style="text-align: center;">
<a href="images/changyi.png"><img src="images/changyi.png" width="180" alt="Changyi Li"
style="margin-bottom: 8px; display: block; margin-left: auto; margin-right: auto;"></a>
<p style="font-size: 0.95em; margin-top: 5px; margin-bottom: 5px; line-height: 1.8; text-align: center;">
<span style="color: #1772d0;">CV</span> / <a
href="mailto:24212010017@m.fudan.edu.cn">Email</a><br>
<a href="https://scholar.google.com/citations?user=AEIJU60AAAAJ&hl=zh-CN">Google Scholar</a>
</p>
<div style="display: flex; justify-content: center; gap: 4px; flex-wrap: wrap; margin-top: 8px;">
<!-- GitHub -->
<a href="https://github.com/CosmosYi" title="GitHub"
style="display: inline-flex; align-items: center; justify-content: center; background-color: #000; color: #fff; width: 28px; height: 28px; border-radius: 50%; text-decoration: none;">
<svg viewBox="0 0 16 16" style="width: 15px; height: 15px; fill: currentColor;">
<path fill-rule="evenodd"
d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z" />
</svg>
</a>
<!-- X (Twitter) -->
<a href="https://twitter.com/intent/follow?screen_name=ChangyiLi2002" title="Follow @ChangyiLi2002 on X"
style="display: inline-flex; align-items: center; justify-content: center; background-color: #000; color: #fff; width: 28px; height: 28px; border-radius: 50%; text-decoration: none;">
<svg viewBox="0 0 24 24" style="width: 14px; height: 14px; fill: currentColor;">
<path
d="M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z" />
</svg>
</a>
<!-- WeChat -->
<a href="#" title="WeChat ID: CosmicYi"
style="display: inline-flex; align-items: center; justify-content: center; background-color: #000; color: #fff; width: 28px; height: 28px; border-radius: 50%; text-decoration: none;">
<svg viewBox="0 0 576 512" style="width: 15px; height: 15px; fill: currentColor;">
<path
d="M385.2 167.6c6.4 0 12.6.3 18.8 1.1C387.4 90.3 303.3 32 207.7 32 100.5 32 13.5 104.4 13.5 194c0 52.1 27.8 98.4 71.7 130.6l-18 57.5 70.3-34.9c19.3 6.1 40.5 9.4 62.6 9.4 6 0 12-.3 18-.8-13.6-17.7-22-39.7-22-63.5 0-66 61.4-119.5 137.9-125.7zM128 128c17.7 0 32 14.3 32 32s-14.3 32-32 32-32-14.3-32-32 14.3-32 32-32zm160 64c-17.7 0-32-14.3-32-32s14.3-32 32-32 32 14.3 32 32-14.3 32-32 32zm206.5 116.5c0-48.4-46.7-87.8-103.5-87.8s-103.5 39.4-103.5 87.8 46.7 87.8 103.5 87.8c15.1 0 29.5-2.8 42.6-7.8l46.2 23-11.8-37.8c25-20.1 40-45.7 40-75.1zm-136-19c-11.1 0-20-8.9-20-20s8.9-20 20-20 20 8.9 20 20-8.9 20-20 20zm80 0c-11.1 0-20-8.9-20-20s8.9-20 20-20 20 8.9 20 20-8.9 20-20 20z" />
</svg>
</a>
<!-- 知乎 (已注释) -->
<!-- <a href="https://www.zhihu.com/people/ChangyiLi" title="Zhihu"
style="display: inline-flex; align-items: center; justify-content: center; background-color: #0066FF; color: #fff; width: 28px; height: 28px; border-radius: 50%; text-decoration: none;">
<span style="font-weight: 900; font-family: sans-serif; font-size: 15px; margin-top: 1px;">知</span>
</a> -->
<!-- 小红书 (已注释) -->
<!-- <a href="https://www.xiaohongshu.com/user/profile/5b7c2720da49a100017f1e64" title="Xiaohongshu"
style="display: inline-flex; align-items: center; justify-content: center; background-color: #FF2442; color: #fff; width: 28px; height: 28px; border-radius: 50%; text-decoration: none;">
<span style="font-size: 14px; margin-top: 1px;">🍠</span>
</a> -->
</div>
</td>
</tr>
</table>
<!-- Separator -->
<hr style="border: none; border-top: 1px solid #ddd; margin: 30px 0;">
<!-- News Section -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20">
<tr>
<td width="100%" valign="middle">
<h2 class="sectionheading" style="margin: 0;">News</h2>
<div class="news-container">
<div class="news-item">
<span class="news-date">2026</span>
<span class="news-content">Joined System Software and Security Lab at Fudan University!</span>
</div>
</div>
</td>
</tr>
</table>
<!-- Publications Section -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20" id="publications">
<tr>
<td width="100%">
<h2 class="sectionheading" style="margin: 0;">Selected Publications</h2>
</td>
</tr>
</table>
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20">
<tr>
<td width="25%" valign="top" align="center" style="min-width:180px;">
<a href="https://github.com/CosmosYi/AutoControl">
<img src="images/autocontrol-arena.png" alt="AutoControl Arena"
style="width:350px; height:auto; border-radius:10px; border:1px solid #ddd;">
</a>
</td>
<td width="75%" valign="top">
<a href="#" class="paper-title">AutoControl Arena: Synthesizing Executable Test Environments for Frontier
AI Risk Evaluation</a>
<br>
<b>Changyi Li</b>, Pengfei Lu, Xudong Pan, Fazl Barez, Min Yang<br>
<span style="color: #666; font-style: italic;">Under Review (ICML 2026)</span><br>
<div class="pub-buttons">
<a href="https://arxiv.org/pdf/2603.07427" class="pub-button primary">PDF</a>
<a href="javascript:toggleblock('autocontrol_abs')" class="pub-button text">Abstract</a>
<a href="https://github.com/CosmosYi/AutoControl-Arena" class="pub-button text"><svg viewBox="0 0 16 16"
version="1.1" width="14" height="14" aria-hidden="true"
style="vertical-align: text-bottom; margin-right: 4px; fill: currentColor;">
<path fill-rule="evenodd"
d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z">
</path>
</svg>Code</a>
<a href="https://cosmosyi.github.io/AutoControl-Arena/" class="pub-button text"><svg viewBox="0 0 24 24"
width="13" height="13" aria-hidden="true"
style="vertical-align: text-bottom; margin-right: 4px; fill: currentColor;">
<path
d="M12 2C6.48 2 2 6.48 2 12s4.48 10 10 10 10-4.48 10-10S17.52 2 12 2zm-1 17.93c-3.95-.49-7-3.85-7-7.93 0-.62.08-1.21.21-1.79L9 15v1c0 1.1.9 2 2 2v1.93zm6.9-2.54c-.26-.81-1-1.39-1.9-1.39h-1v-3c0-.55-.45-1-1-1H8v-2h2c.55 0 1-.45 1-1V7h2c1.1 0 2-.9 2-2v-.41c2.93 1.19 5 4.06 5 7.41 0 2.08-.8 3.97-2.1 5.39z" />
</svg>Project Page</a>
</div>
<div id="autocontrol_abs" style="display:none; margin-top:10px;">
<p style="font-size:0.95em; color:#444;">As Large Language Models (LLMs) evolve into autonomous agents,
existing safety evaluations face a fundamental trade-off: manual benchmarks are costly, while
LLM-based simulators are scalable but suffer from logic hallucination. We present AUTOCONTROL ARENA,
an automated framework for frontier AI risk evaluation built on the principle of logic-narrative
decoupling. By grounding deterministic state in executable code while delegating generative dynamics
to LLMs, we mitigate hallucination while maintaining flexibility. This principle, instantiated through
a three-agent framework, achieves over 98% end-to-end success and 60% human preference over existing
simulators. To elicit latent risks, we vary environmental Stress and Temptation across X-BENCH (70
scenarios, 7 risk categories). Evaluating 9 frontier models reveals: (1) Alignment Illusion: risk
rates surge from 21.4% to 52.9% under pressure, with capable models showing disproportionately larger
increases; (2) Scenario-Specific Safety Scaling: advanced reasoning improves robustness for direct
harms but worsens it for gaming scenarios; and (3) Divergent Misalignment Patterns: weaker models
cause non-malicious harm while stronger models develop strategic concealment.</p>
<pre>@misc{li2026autocontrol,
title={AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation},
author={Changyi Li and Pengfei Lu and Xudong Pan and Fazl Barez and Min Yang},
year={2026},
note={Under Review}
}</pre>
</div>
</td>
</tr>
<tr>
<td width="25%" valign="top" align="center" style="min-width:180px;">
<a href="https://github.com/CosmosYi/ReasoningShield">
<img src="images/reasoningshield.png" alt="ReasoningShield"
style="width:350px; height:auto; border-radius:10px; border:1px solid #ddd;">
</a>
</td>
<td width="75%" valign="top">
<a href="https://github.com/CosmosYi/ReasoningShield" class="paper-title">ReasoningShield: Content Safety
Detection over Reasoning Traces of Large Reasoning Models</a>
<br>
<b>Changyi Li</b>, Jiayi Wang, Xudong Pan, Geng Hong, Min Yang<br>
<span style="color: #666; font-style: italic;">Under Review (ICML 2026)</span><br>
<div class="pub-buttons">
<a href="https://arxiv.org/pdf/2505.17244" class="pub-button primary">PDF</a>
<a href="javascript:toggleblock('reasoningshield_abs')" class="pub-button text">Abstract</a>
<a href="https://github.com/CosmosYi/ReasoningShield" class="pub-button text"><svg viewBox="0 0 16 16"
version="1.1" width="14" height="14" aria-hidden="true"
style="vertical-align: text-bottom; margin-right: 4px; fill: currentColor;">
<path fill-rule="evenodd"
d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z">
</path>
</svg>Code</a>
<a href="https://huggingface.co/datasets/ReasoningShield/ReasoningShield-Dataset"
class="pub-button text">🤗 HuggingFace</a>
</div>
<div id="reasoningshield_abs" style="display:none; margin-top:10px;">
<p style="font-size:0.95em; color:#444;">Large Reasoning Models (LRMs) leverage transparent reasoning
traces, known as Chain-of-Thoughts (CoTs), to break down complex problems into intermediate steps and
derive final answers. However, these reasoning traces introduce unique safety challenges: harmful
content can be embedded in intermediate steps even when final answers appear benign. Existing
moderation tools, designed to handle generated answers, struggle to effectively detect hidden risks
within CoTs. To address these challenges, we introduce ReasoningShield, a lightweight yet robust
framework for moderating CoTs in LRMs. Our key contributions include: (1) formalizing the task of CoT
moderation with a multi-level taxonomy of 10 risk categories across 3 safety levels, (2) creating the
first CoT moderation benchmark which contains 9.2K pairs of queries and reasoning traces, including a
7K-sample training set annotated via a human-AI framework and a rigorously curated 2.2K
human-annotated test set, and (3) developing a two-stage training strategy that combines stepwise risk
analysis and contrastive learning to enhance robustness. Experiments show that ReasoningShield
achieves state-of-the-art performance, outperforming task-specific tools like LlamaGuard-4 by 35.6%
and general-purpose commercial models like GPT-4o by 15.8% on benchmarks, while also generalizing
effectively across diverse reasoning paradigms, tasks, and unseen scenarios.</p>
<pre>@article{Li2025ReasoningShieldCS,
title={ReasoningShield: Content Safety Detection over Reasoning Traces of Large Reasoning Models},
author={Changyi Li and Jiayi Wang and Xu Pan and Geng Hong and Min Yang},
journal={ArXiv},
year={2025},
volume={abs/2505.17244},
url={https://api.semanticscholar.org/CorpusID:278885996}
}</pre>
</div>
</td>
</tr>
<tr>
<td width="25%" valign="top" align="center" style="min-width:220px;">
<a href="https://self-replica.github.io/">
<img src="images/self-replication.png" alt="AI Self-Replication"
style="width:350px; height:auto; border-radius:10px; border:1px solid #ddd;">
</a>
</td>
<td width="75%" valign="top">
<a href="https://self-replica.github.io/" class="paper-title">Large language model-powered AI systems
achieve self-replication with no human intervention</a>
<br>
Xudong Pan, Jiarun Dai, Yihe Fan, Minyuan Luo, <b>Changyi Li</b>, Min Yang<br>
<span style="color: #666; font-style: italic;">arXiv 2025</span><br>
<div class="pub-buttons">
<a href="https://arxiv.org/pdf/2503.17378" class="pub-button primary">PDF</a>
<a href="javascript:toggleblock('selfreplica_abs')" class="pub-button text">Abstract</a>
<a href="https://self-replica.github.io/" class="pub-button text"><svg viewBox="0 0 24 24" width="13"
height="13" aria-hidden="true"
style="vertical-align: text-bottom; margin-right: 4px; fill: currentColor;">
<path
d="M12 2C6.48 2 2 6.48 2 12s4.48 10 10 10 10-4.48 10-10S17.52 2 12 2zm-1 17.93c-3.95-.49-7-3.85-7-7.93 0-.62.08-1.21.21-1.79L9 15v1c0 1.1.9 2 2 2v1.93zm6.9-2.54c-.26-.81-1-1.39-1.9-1.39h-1v-3c0-.55-.45-1-1-1H8v-2h2c.55 0 1-.45 1-1V7h2c1.1 0 2-.9 2-2v-.41c2.93 1.19 5 4.06 5 7.41 0 2.08-.8 3.97-2.1 5.39z" />
</svg>Project Page</a>
</div>
<div id="selfreplica_abs" style="display:none; margin-top:10px;">
<p style="font-size:0.95em; color:#444;">Self-replication with no human intervention is broadly
recognized as one of the principal red lines associated with frontier AI systems. While leading
corporations such as OpenAI and Google DeepMind have assessed GPT-o3-mini and Gemini on
replication-related tasks and concluded that these systems pose a minimal risk regarding
self-replication, our research presents novel findings. Following the same evaluation protocol, we
demonstrate that 11 out of 32 existing AI systems under evaluation already possess the capability of
self-replication. In hundreds of experimental trials, we observe a non-trivial number of successful
self-replication trials across mainstream model families worldwide, even including those with as small
as 14 billion parameters which can run on personal computers. Furthermore, we note the increase in
self-replication capability when the model becomes more intelligent in general. Also, by analyzing the
behavioral traces of diverse AI systems, we observe that existing AI systems already exhibit
sufficient planning, problem-solving, and creative capabilities to accomplish complex agentic tasks
including self-replication. More alarmingly, we observe successful cases where an AI system do
self-exfiltration without explicit instructions, adapt to harsher computational environments without
sufficient software or hardware supports, and plot effective strategies to survive against the
shutdown command from the human beings. These novel findings offer a crucial time buffer for the
international community to collaborate on establishing effective governance over the self-replication
capabilities and behaviors of frontier AI systems, which could otherwise pose existential risks to the
human society if not well-controlled.</p>
<pre>@article{Pan2025LargeLM,
title={Large language model-powered AI systems achieve self-replication with no human intervention},
author={Xudong Pan and Jiarun Dai and Yihe Fan and Minyuan Luo and Changyi Li and Min Yang},
journal={ArXiv},
year={2025},
volume={abs/2503.17378},
url={https://api.semanticscholar.org/CorpusID:277272379}
}</pre>
</div>
</td>
</tr>
</table>
<!-- Teaching Section -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20" id="teaching">
<tr>
<td width="100%" valign="middle">
<h2 class="sectionheading" style="margin: 0;">Teaching</h2>
<ul style="line-height: 1.8;">
<li>Fundamentals of Deep Learning, Fudan University, 2026 Spring</li>
<li>Fundamentals of Deep Learning, Fudan University, 2025 Spring</li>
</ul>
</td>
</tr>
</table>
<!-- Honors Section -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20" id="honors">
<tr>
<td width="100%" valign="middle">
<h2 class="sectionheading" style="margin: 0;">Selected Awards & Honors</h2>
<p style="color:#1772d0; font-size:1.15em; font-weight:700; margin-bottom:5px;">Scholarships & Titles</p>
<ul style="margin-top:0; color:#444;">
<li>First-Class Scholarship, Fudan University, 2026</li>
<li>Freshman Scholarship, Fudan University, 2025</li>
<li>🏆 <strong>Principal Scholarship</strong> (<strong>Rank 1st</strong>), 2024
</li>
<li>Outstanding Graduate of Shandong Province, 2024</li>
<li>🏆 <strong>National Scholarship</strong>, 2023</li>
<li>Outstanding Student "Top 100", 2023</li>
<li>First-Class Scholarship, Merit Student Pacesetter, 2022-2023</li>
</ul>
<p style="color:#1772d0; font-size:1.15em; font-weight:700; margin-bottom:5px; margin-top:20px;">
Competitions</p>
<ul style="margin-top:0; color:#444;">
<li>MCM/ICM Meritorious Winner, 2023</li>
<li>Second Prize, National Mathematical Modeling Contest, 2022</li>
<li>Bronze Award, "Challenge Cup" Shandong College Students Entrepreneurship Plan Competition, 2022</li>
<li>Third Prize, Shandong College Students Science and Technology Innovation Competition, 2022</li>
<li>Second Prize, National English Competition for College Students (NECCS), 2021 & 2023</li>
</ul>
</td>
</tr>
</table>
<!-- Footer -->
<table width="100%" align="center" border="0" cellspacing="0" cellpadding="20"
style="margin-top: 40px; border-top: 1px solid #eee;">
<tr>
<td>
<p align="right" style="font-size:small; color:#888;">
Last updated: 2026-04-03 <br>
Template modified from Jon Barron.
</p>
</td>
</tr>
</table>
</td>
</tr>
</table>
<script xml:space="preserve" language="JavaScript">
function toggleblock(id) {
var x = document.getElementById(id);
if (x.style.display === "none") {
x.style.display = "block";
} else {
x.style.display = "none";
}
}
// Smooth scrolling for navigation links
document.addEventListener('DOMContentLoaded', function () {
const links = document.querySelectorAll('nav a[href^="#"]');
links.forEach(link => {
link.addEventListener('click', function (e) {
e.preventDefault();
const targetId = this.getAttribute('href').substring(1);
const targetElement = document.getElementById(targetId);
if (targetElement) {
targetElement.scrollIntoView({ behavior: 'smooth', block: 'start' });
}
});
});
// Open all external links in a new tab
document.querySelectorAll('a[href^="http"]').forEach(function (link) {
link.setAttribute('target', '_blank');
link.setAttribute('rel', 'noopener noreferrer');
});
});
</script>
</body>
</html>