-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy path2.2_responsibility_part2.html
More file actions
530 lines (454 loc) · 30.9 KB
/
2.2_responsibility_part2.html
File metadata and controls
530 lines (454 loc) · 30.9 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>2.2 AI system security vulnerabilities and attacks - Responsibility</title>
<link href="https://fonts.googleapis.com/css2?family=Figtree:wght@300;400;500;600;700&display=swap" rel="stylesheet">
<style>
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
body {
font-family: 'Figtree', -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif;
background-color: #ffffff;
color: #000000;
line-height: 1.3;
}
.container {
max-width: 1200px;
margin: 0 auto;
padding: 8px;
flex: 1;
min-width: 200px;
overflow-wrap: break-word;
word-break: break-word; }
h1 {
text-align: center;
margin-bottom: 8px;
color: #000000;
font-weight: 600;
font-size: 18px;
}
.legend {
text-align: center;
font-size: 12px;
color: #888888;
font-style: italic;
margin-bottom: 12px;
padding: 8px;
background-color: #f9f9f9;
border-radius: 5px;
border: 1px solid #e0e0e0;
}
.selection-title {
text-align: center;
font-size: 14px;
font-weight: 600;
color: #666666;
margin-bottom: 10px;
}
.nav-pills {
display: flex;
flex-wrap: wrap;
gap: 4px;
margin-bottom: 15px;
justify-content: center;
}
.nav-pill {
background: #f8f9fa;
border: 1px solid #e0e0e0;
border-radius: 25px;
padding: 12px 20px;
cursor: pointer;
font-family: 'Figtree', sans-serif;
font-size: 16px;
font-weight: 500;
transition: all 0.3s ease;
color: #000000;
}
.nav-pill:hover {
background: #e9ecef;
border-color: #000000;
}
.nav-pill.active {
background: #000000;
color: white;
border-color: #000000;
}
.actor-section {
display: none;
}
.actor-section.active {
display: block;
}
.content-grid {
display: flex;
width: 100%;
gap: 4px;
}
.content-column {
background: #ffffff;
border: 1px solid #e0e0e0;
border-radius: 8px;
padding: 8px;
flex: 1;
min-width: 200px;
overflow-wrap: break-word;
word-break: break-word; }
.criteria-header {
font-size: 12px;
font-weight: 600;
margin-bottom: 15px;
padding-bottom: 10px;
border-bottom: 2px solid;
}
.criteria-header.higher {
color: #FF0000;
border-bottom-color: #FF0000;
}
.criteria-header.lower {
color: #2E5C8A;
border-bottom-color: #2E5C8A;
}
.summary-section {
margin-bottom: 20px;
}
.summary-text {
margin-bottom: 15px;
font-weight: 500;
color: #000000;
font-size: 15px;
}
.quote-details {
margin-top: 15px;
}
.quote-toggle {
cursor: pointer;
color: #000000;
font-weight: 500;
font-size: 16px;
background-color: #ffff00;
padding: 10px 15px;
border-radius: 4px;
display: inline-block;
}
.quote-toggle:hover {
color: #333333;
}
.quote-list {
margin-top: 15px;
padding-left: 20px;
}
.quote-list li {
margin-bottom: 12px;
font-size: 16px;
padding: 10px 15px;
line-height: 1.3;
color: #000000;
}
@media (max-width: 768px) {
.content-grid {
gap: 4px;
}
.selection-title {
text-align: center;
font-size: 14px;
font-weight: 600;
color: #666666;
margin-bottom: 10px;
}
.nav-pills {
justify-content: flex-start;
}
.nav-pill {
font-size: 16px;
padding: 4px 8px;
}
}
</style>
</head>
<body>
<div class="container">
<h1>2.2 AI system security vulnerabilities and attacks - Responsibility</h1>
<div class="selection-title">Select an actor:</div>
<div class="nav-pills">
<button class="nav-pill active" data-target="AIDeveloperSpecializedAI">
AI Developer (Specialized AI)
</button>
<button class="nav-pill" data-target="AIInfrastructureProvider">
AI Infrastructure Provider
</button>
<button class="nav-pill" data-target="AffectedStakeholder">
Affected Stakeholder
</button>
</div>
<div class="content-sections">
<div class="actor-section active" id="AIDeveloperSpecializedAI">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Respondents emphasized high-to-primary responsibility because specialized developers create domain-specific models for sensitive applications (healthcare, finance), with control over model design and data selection. Their decisions can introduce or mitigate vulnerabilities, giving them high responsibility. They often also deploy systems in sensitive domains (health/finance/OT) that carry higher stakes. They argued that due to damage suffered by systems, specialized developers should bear primary responsibility.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (3)</summary>
<ul class="quote-list">
<li>"Due to the damage suffered by the system, I believe that the developers of the system (AI Developer, AI Developer-Specialized AI) should bear the primary responsibility, so they are the primary response. And AI Developer General Purpose AI mainly provides model services, and they should be highly responsible to system attacks."</li> <li>"I strongly agree with the second half of choices. the question to note is if the affected stakeholder had the choice of being effected by AI or not. It is odd that folks would place primary responsibility on the generalized AI developers because the overall architecture should not be responsible for the specialized actions undergone. It would be the specialized AI tools with their claims of effectiveness that would warrant the higher scrutiny. It would seem similar to those who design a general purpose economic model being scrutinized because it did not apply to a specific use-case."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
Specialized devs = primary: often also deploy; domain systems (health/finance/OT) carry higher stakes."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> [NO EXPERT COMMENTS PROVIDED]</p>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIInfrastructureProvider">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> AI developers and third-party vendors/cloud providers carry very high responsibility, as they possess the technical expertise to implement security safeguards, control critical components of AI systems, and directly influence whether vulnerabilities exist or are mitigated.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (2)</summary>
<ul class="quote-list">
<li>"AI developers and third-party vendors/cloud providers carry very high responsibility, as they possess the technical expertise to implement security safeguards, control critical components of AI systems, and directly influence whether vulnerabilities exist or are mitigated."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
GP devs & infra = high: must ship secure artifacts (signed/attested weights, supply-chain hygiene), poisoning defenses, and strong multi-tenant isolation/KMS/observability controls."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Multiple respondents argued fpr minimal responsibility because infrastructure providers don't acquire training data, build models, train models, or deploy models. They only provide underlying hardware and infrastructure, with primary responsibility being to ensure resources are properly protected.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"Again, I see the AI Infrastructure providers as low responsibility. They must have secure infrastructure and avoid unintended sharing. But they do not acquire the training data, build the models, train the models, or deploy the models. They should be at most minimally responsible for what organizations using their infrastructure do with it."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AffectedStakeholder">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> [SUMMARY TBC]</p>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> One expert commented: "Responsibility follows who controls the attack surface and can remediate fast. affected stakeholders = none: they can practice hygiene but don't control architecture or patch cadence."</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"Responsibility follows who controls the attack surface and can remediate fast.
Users = minimal; affected stakeholders = none: they can practice hygiene but don't control architecture or patch cadence."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIDeveloperSpecializedAI">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Respondents emphasized high-to-primary responsibility because specialized developers create domain-specific models for sensitive applications (healthcare, finance), with control over model design and data selection. Their decisions can introduce or mitigate vulnerabilities, giving them high responsibility. They often also deploy systems in sensitive domains (health/finance/OT) that carry higher stakes. They argued that due to damage suffered by systems, specialized developers should bear primary responsibility.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (3)</summary>
<ul class="quote-list">
<li>"Due to the damage suffered by the system, I believe that the developers of the system (AI Developer, AI Developer-Specialized AI) should bear the primary responsibility, so they are the primary response. And AI Developer General Purpose AI mainly provides model services, and they should be highly responsible to system attacks."</li> <li>"I strongly agree with the second half of choices. the question to note is if the affected stakeholder had the choice of being effected by AI or not. It is odd that folks would place primary responsibility on the generalized AI developers because the overall architecture should not be responsible for the specialized actions undergone. It would be the specialized AI tools with their claims of effectiveness that would warrant the higher scrutiny. It would seem similar to those who design a general purpose economic model being scrutinized because it did not apply to a specific use-case."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
Specialized devs = primary: often also deploy; domain systems (health/finance/OT) carry higher stakes."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> [NO EXPERT COMMENTS PROVIDED]</p>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIDeployer">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Their deployment choices directly affect exposure to security risks. They also bear high responsibility given their role in integrating AI into operational environments and maintaining system security—misconfigurations or weak monitoring substantially amplify risks. Deployers run the live stack and decide hardening, patching, and incident response, making them primarily responsible. Responsibility follows who controls the attack surface and can remediate fast.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (2)</summary>
<ul class="quote-list">
<li>"AI deployers and organizations also bear high to very high responsibility, given their role in integrating AI into operational environments and maintaining system security; misconfigurations or weak monitoring can substantially amplify risks."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
Deployers = primary: they run the live stack (models, RAG, agents/plugins, CI/CD, secrets, logging) and decide hardening, patching, and incident response."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> One expert argued "The model is not where the vulnerability is - it is in the AI system. Model deployment must engage in security posture management that mitigates known and unknown vulnerabilities. Putting the responsibility with model developers is wrong-headed, as no system is expected to secure itself (thx Godel!). Model developers may deploy their models (especially as services), in which case they are a Deployer. But otherwise the consensus view that Model Developers are primarily responsible for security vulnerabilities doesn't reflect reality. It's even difficult to hold software vendors responsible for software vulnerabilities in the current software ecosystem."</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"The model is not where the vulnerability is - it is in the AI system. Model deployment must engage in security posture management that mitigates known and unknown vulnerabilities. Putting the responsibility with model developers is wrong-headed, as no system is expected to secure itself (thx Godel!). Model developers may deploy their models (especially as services), in which case they are a Deployer. But otherwise the consensus view that Model Developers are primarily responsible for security vulnerabilities doesn't reflect reality. It's even difficult to hold software vendors responsible for software vulnerabilities in the current software ecosystem."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIInfrastructureProvider">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> AI developers and third-party vendors/cloud providers carry very high responsibility, as they possess the technical expertise to implement security safeguards, control critical components of AI systems, and directly influence whether vulnerabilities exist or are mitigated.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (2)</summary>
<ul class="quote-list">
<li>"AI developers and third-party vendors/cloud providers carry very high responsibility, as they possess the technical expertise to implement security safeguards, control critical components of AI systems, and directly influence whether vulnerabilities exist or are mitigated."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
GP devs & infra = high: must ship secure artifacts (signed/attested weights, supply-chain hygiene), poisoning defenses, and strong multi-tenant isolation/KMS/observability controls."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Multiple respondents argued fpr minimal responsibility because infrastructure providers don't acquire training data, build models, train models, or deploy models. They only provide underlying hardware and infrastructure, with primary responsibility being to ensure resources are properly protected.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"Again, I see the AI Infrastructure providers as low responsibility. They must have secure infrastructure and avoid unintended sharing. But they do not acquire the training data, build the models, train the models, or deploy the models. They should be at most minimally responsible for what organizations using their infrastructure do with it."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIUser">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> One respondent argued "AI users can include any software engineers building systems with generated code that may contain security vulnerabilities. They bear a decent amount of responsibility for reviewing that output." Another mentioned that they "can contribute to risk reduction through basic security hygiene and cautious behavior"</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"AI users can include any software engineers building systems with generated code that may contain security vulnerabilities. They bear a decent amount of responsibility for reviewing that output."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Users are at most minimally responsible due to minimal expected capacity (expertise, resources) to effectively control against attacks. Putting onus on users would reduce accountability of companies, going against lessons learned throughout history of product safety, engineering, and regulation. Users generally lack technical capacity to prevent advanced attacks, and they don't control architecture or patch cadence.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (3)</summary>
<ul class="quote-list">
<li>"End users and consumers have low to moderate responsibility, as they can contribute to risk reduction through basic security hygiene and cautious behavior, but generally lack the technical capacity to prevent advanced attacks."</li> <li>"The AI user is at most minimally responsible, due to their minimal expected capacity (i.e., expertise, resources) to effectively control against attacks. Putting the onus on the user would reduce the accountability of companies, going against the lessons learned throughout the history of product safety, engineering, regulation, etc."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
Users = minimal; affected stakeholders = none: they can practice hygiene but don't control architecture or patch cadence."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AffectedStakeholder">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> [SUMMARY TBC]</p>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> One expert commented: "Responsibility follows who controls the attack surface and can remediate fast. affected stakeholders = none: they can practice hygiene but don't control architecture or patch cadence."</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (1)</summary>
<ul class="quote-list">
<li>"Responsibility follows who controls the attack surface and can remediate fast.
Users = minimal; affected stakeholders = none: they can practice hygiene but don't control architecture or patch cadence."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
<div class="actor-section" id="AIGovernanceActor">
<div class="content-grid">
<div class="content-column">
<h3 class="criteria-header higher">Reasons for Higher Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Multiple respondents emphasized moderate-to-high responsibility because governance actors set policies, standards, and oversight mechanisms. They influence risk mitigation through regulation and accountability frameworks, shaping the environment even without direct technical control. They establish standards, enforce compliance, and shape sector-wide security practices. They can make controls mandatory including secure SDLC, artifact provenance, and reporting/audit requirements. They can set enforceable standards, audit compliance, and coordinate international regulatory responses.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (2)</summary>
<ul class="quote-list">
<li>"Regulators and government agencies hold moderate to high responsibility, as they establish standards, enforce compliance, and shape sector-wide security practices, though their direct technical influence is limited."</li> <li>"Responsibility follows who controls the attack surface and can remediate fast.
Governance = high: make these controls mandatory (secure SDLC, artifact provenance, reporting/audit)."</li>
</ul>
</details>
</div>
</div>
<div class="content-column">
<h3 class="criteria-header lower">Reasons for Lower Responsibility</h3>
<div class="summary-section">
<p class="summary-text"><strong>AI-generated summary:</strong> Some argued governance actors have minimal-to-moderate responsibility, arguing that saying governance actors are responsible for AI risks is like saying judges are responsible for crimes. They claim governance actors aren't really in a position to address this risk directly, so while responsible for building the ecosystem, it doesn't make sense to consider them responsible for this specific risk.</p>
<details class="quote-details">
<summary class="quote-toggle">See all expert comments (2)</summary>
<ul class="quote-list">
<li>"I continue to assess that AI governance actors are "minimally responsible". I think there is a category error at work here. Saying AI governance actors are responsible for AI risks is like saying that judges are responsible for crimes being committed. The kind of responsibility a judge has is very different from the kind of responsibility that a criminal or a lock-pick-maker has. The better way to think of this is that the AI governance actor is responsible for holding responsible the actor who is properly responsible. It would be recursive that AI governance actor is themselves responsible. Would we propose some meta-AI-goverance governor who holds responsible the AI governance actors that fail to hold responsible the actors that should be responsible? This is not the right way of thinking."</li> <li>"Governance actors aren't really in a position to address this risk directly, so while they are responsible for building the ecosystem, I don't think it makes sense to consider them 'responsible' for this risk specifically."</li>
</ul>
</details>
</div>
</div>
</div>
</div>
</div>
</div>
<script>
document.addEventListener('DOMContentLoaded', function() {
const pills = document.querySelectorAll('.nav-pill');
const sections = document.querySelectorAll('.actor-section');
pills.forEach(pill => {
pill.addEventListener('click', function() {
// Remove active class from all pills and sections
pills.forEach(p => p.classList.remove('active'));
sections.forEach(s => s.classList.remove('active'));
// Add active class to clicked pill
this.classList.add('active');
// Show corresponding section
const targetId = this.getAttribute('data-target');
const targetSection = document.getElementById(targetId);
if (targetSection) {
targetSection.classList.add('active');
}
});
});
});
</script>
</body>
</html>