-
Notifications
You must be signed in to change notification settings - Fork 324
/
aoai_checklist.en.json
1070 lines (1070 loc) · 52.9 KB
/
aoai_checklist.en.json
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
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
{
"items": [
{
"category": "Responsible AI",
"subcategory": "Metaprompting",
"text": "Follow Metaprompting guardrails for resonsible AI",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "a85b86ad-884f-48e3-9273-4b875ba18f10",
"id": "AOAI.1",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/concepts/system-message#define-additional-safety-and-behavioral-guardrails"
},
{
"category": "Operations Management",
"subcategory": "Load Balancing",
"text": "Consider Gateway patterns with APIM or solutions like AI central for better rate limiting, load balancing, authentication and logging",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "d4391898-cd28-48be-b6b1-7cb8245451e1",
"id": "AOAI.10",
"severity": "High",
"link": "https://github.com/Azure-Samples/AI-Gateway"
},
{
"category": "Operations Management",
"subcategory": "Monitoring",
"text": "Enable monitoring for your AOAI instances",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "aed3453a-ec72-4392-97a1-52d6cc5e4029",
"id": "AOAI.11",
"severity": "High",
"link": "https://techcommunity.microsoft.com/t5/fasttrack-for-azure/azure-openai-insights-monitoring-ai-with-confidence/ba-p/4026850"
},
{
"category": "Operations Management",
"subcategory": "Alerts",
"text": "Create alerts to notify teams of events such as an entry in the activity log created by an action performed on the resource, such as regenerating its subscription keys or a metric threshold such as the number of errors exceeding 10 in an hour",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "697cb391-ed16-4b2d-886f-0a0241addde6",
"id": "AOAI.12",
"graph": "resources | where type == 'microsoft.insights/metricalerts' | extend compliant = (properties.targetResourceType =~ 'Microsoft.CognitiveServices/accounts') | project id, compliant",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/monitoring#set-up-alerts"
},
{
"category": "Operations Management",
"subcategory": "Monitoring",
"text": "Monitor token usage to prevent service disruptions due to capacity",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "8a477cde-b486-41bc-9bc1-0ae66e25d4d5",
"id": "AOAI.13",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/monitoring"
},
{
"category": "Operations Management",
"subcategory": "Observability",
"text": "observe metrics like processed inference tokens, generated completion tokens monitor for rate limit",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "a3aec2c4-e243-46b0-936c-b45e17960eee",
"id": "AOAI.14",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/monitoring"
},
{
"category": "Operations Management",
"subcategory": "Observability",
"text": "Enable and configure Diagnostics for the Azure OpenAI Service. If not sufficient, consider using a gateway such as Azure API Managements in front of Azure OpenAI to log both incoming prompts and outgoing responses, where permitted",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "fbdf4cc2-eec4-4d76-8c31-d25ffbb46a39",
"id": "AOAI.15",
"severity": "Low",
"link": "https://techcommunity.microsoft.com/t5/apps-on-azure-blog/build-an-enterprise-ready-azure-openai-solution-with-azure-api/ba-p/3907562"
},
{
"category": "Operations Management",
"subcategory": "Infrastructure Deployment",
"text": "Use Infrastructure as code to deploy the Azure OpenAI Service, model deployments, and all related resources",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "3af30ed3-2947-498b-8178-a2c5a46ceb54",
"id": "AOAI.16",
"severity": "High",
"link": "https://github.com/Azure-Samples/openai-enterprise-iac"
},
{
"category": "Governance and Security",
"subcategory": "Authentication",
"text": "Use Microsoft Entra Authentication with Managed Identity instead of API Key",
"waf": "Security",
"service": "OpenAI",
"guid": "4350d092-d234-4292-a752-8537a551c5bf",
"id": "AOAI.17",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/managed-identity"
},
{
"category": "Responsible AI",
"subcategory": "Evaluation",
"text": "Evaluate the performance/accuracy of the system with a known golden dataset which has the inputs and the correct answers. Leverage capabilities in PromptFlow for Evaluation.",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "4e4f1854-287d-45cd-a126-cc031af5b1fc",
"id": "AOAI.18",
"severity": "High",
"link": "https://learn.microsoft.com/azure/machine-learning/prompt-flow/how-to-bulk-test-evaluate-flow?view=azureml-api-2"
},
{
"category": "Operations Management",
"subcategory": "Hosting model",
"text": "Evaluate usage of Provisioned throughput model ",
"waf": "Performance",
"service": "OpenAI",
"guid": "68889535-e327-4897-b31b-67d67be5962a",
"id": "AOAI.19",
"severity": "High",
"link": "https://learn.microsoft.com/azure/architecture/ai-ml/architecture/baseline-openai-e2e-chat#azure-openai---performance-efficiency"
},
{
"category": "Responsible AI",
"subcategory": "Content Safety",
"text": "Review and implement Azure AI content safety",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "cd288bed-6b17-4cb8-8454-51e1aed3453a",
"id": "AOAI.2",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/content-safety/overview"
},
{
"category": "Operations Management",
"subcategory": "Throughput definition",
"text": "Define and evaluate the throughput of the system based on tokens & response per minute and align with requirements",
"waf": "Performance",
"service": "OpenAI",
"guid": "1193846d-697c-4b39-8ed1-6b2d186f0a02",
"id": "AOAI.20",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/latency#system-level-throughput"
},
{
"category": "Operations Management",
"subcategory": "Latency improvement",
"text": "Improve latency of the system by limiting token sizes, streaming options for applications like chatbots or conversational interfaces. Streaming can enhance the perceived performance of Azure OpenAI applications by delivering responses to users in an incremental manner",
"waf": "Performance",
"service": "OpenAI",
"guid": "41addde6-8a47-47cd-bb48-61bc3bc10ae6",
"id": "AOAI.21",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/latency#improve-performance"
},
{
"category": "Operations Management",
"subcategory": "Elasticity segregation",
"text": "Estimate elasticity demands to determine synchronous and batch request segregation based on priority. For high priority, use synchronous approach and for low priority, asynchronous batch processing with queue is preferred",
"waf": "Performance",
"service": "OpenAI",
"guid": "6e25d4d5-a3ae-4c2c-9e24-36b0336cb45e",
"id": "AOAI.22",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/latency#batching"
},
{
"category": "Operations Management",
"subcategory": "Benchmarking",
"text": "Benchmark token consumption requirements based on estimated demands from consumers. Consider using the Azure OpenAI benchmarking tool to help you validate the throughput if you are using Provisioned Throughput Unit deployments",
"waf": "Performance",
"service": "OpenAI",
"guid": "5bda4332-4f24-4811-9331-82ba51752694",
"id": "AOAI.23",
"severity": "High",
"link": "https://github.com/Azure/azure-openai-benchmark/"
},
{
"category": "Operations Management",
"subcategory": "Elasticity ",
"text": "If you are using Provisioned Throughput Units (PTUs), consider deploying a token-per-minute (TPM) deployment for overflow requests. Use a gateway to route requests to the TPM deployment when the PTU limits are reached.",
"waf": "Performance",
"service": "OpenAI",
"guid": "4008ae7d-7e47-4432-96d8-bdcf55bce619",
"id": "AOAI.24",
"severity": "Medium",
"link": "https://techcommunity.microsoft.com/t5/fasttrack-for-azure/optimizing-azure-openai-a-guide-to-limits-quotas-and-best/ba-p/4076268"
},
{
"category": "Operations Management",
"subcategory": "Model choice",
"text": "Choose the right model for the right task. Pick models with right tradeoff between speed, quality of response and output complexity",
"waf": "Performance",
"service": "OpenAI",
"guid": "e8a13f98-8794-424d-9267-86d60b96c97b",
"id": "AOAI.25",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/concepts/models"
},
{
"category": "Operations Management",
"subcategory": "Fine tuning",
"text": "Have a baseline for performance without fine-tuning for knowing whether or not fine-tuning has improved model performance",
"waf": "Performance",
"service": "OpenAI",
"guid": "e9951904-8384-45c9-a6cb-2912156a1147",
"id": "AOAI.26",
"severity": "Medium",
"link": "https://github.com/Azure/azure-openai-benchmark/"
},
{
"category": "BC and DR",
"subcategory": "Multi-region architecture",
"text": "Deploy multiple OAI instances across regions",
"waf": "Reliability",
"service": "OpenAI",
"guid": "5e39f541-accc-4d97-a376-bcdb3750ab2a",
"id": "AOAI.27",
"severity": "Low",
"link": "https://learn.microsoft.com/azure/architecture/ai-ml/architecture/baseline-openai-e2e-chat#azure-openai---reliability"
},
{
"category": "BC and DR",
"subcategory": "Load balancing",
"text": "Implement retry & healthchecks with Gateway pattern like APIM",
"waf": "Reliability",
"service": "OpenAI",
"guid": "b039da6d-55d7-4c89-8adb-107d5325af62",
"id": "AOAI.28",
"severity": "High",
"link": "https://learn.microsoft.com/azure/architecture/ai-ml/architecture/baseline-openai-e2e-chat#azure-openai---reliability"
},
{
"category": "BC and DR",
"subcategory": "Quotas",
"text": "Ensure having adequate quotas of TPM & RPM for the workload",
"waf": "Reliability",
"service": "OpenAI",
"guid": "5ca44e46-85e2-4223-ace8-bb12308ca5f1",
"id": "AOAI.29",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/quota?tabs=rest#introduction-to-quota"
},
{
"category": "Responsible AI",
"subcategory": "UX best practice",
"text": "Review the considerations in HAI toolkit guidance and apply those interaction practices for the slution",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "ec723923-7a15-42d6-ac5e-402925387e5c",
"id": "AOAI.3",
"severity": "Medium",
"link": "https://www.microsoft.com/research/project/guidelines-for-human-ai-interaction/"
},
{
"category": "BC and DR",
"subcategory": "Load balancing",
"text": "Deploy separate fine tuned models across regions if finetuning is employed",
"waf": "Reliability",
"service": "OpenAI",
"guid": "7f154e3a-a369-4282-ae7e-316183687a04",
"id": "AOAI.30",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/business-continuity-disaster-recovery"
},
{
"category": "BC and DR",
"subcategory": "Data Backup and Disaster Recovery",
"text": "Regularly backup and replicate critical data to ensure data availability and recoverability in case of data loss or system failures. Leverage Azure's backup and disaster recovery services to protect your data.",
"waf": "Reliability",
"service": "OpenAI",
"guid": "77a1f893-5bda-4433-84f2-4811633182ba",
"id": "AOAI.31",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/backup/backup-overview"
},
{
"category": "BC and DR",
"subcategory": "SLA considerations",
"text": "Azure AI search service tiers should be choosen to have a SLA ",
"waf": "Reliability",
"service": "OpenAI",
"guid": "95b96ad8-844c-4e3b-8b38-b876ba2cf204",
"id": "AOAI.32",
"graph": "resources | where type == 'microsoft.search/searchservices' | extend compliant = (sku.name != 'free' and properties.replicaCount >= 3) | project id, compliant",
"severity": "High",
"link": "https://learn.microsoft.com/azure/search/search-reliability"
},
{
"category": "Governance and Security",
"subcategory": "Data Sensitivity",
"text": "Classify data and sensitivity, labeling with Microsoft Purview before generating the embeddings and make sure to treat the embeddings generated with same sensitivity and classification",
"waf": "Security",
"service": "OpenAI",
"guid": "99013a5d-3ce4-474d-acbd-8682a6abca2a",
"id": "AOAI.33",
"severity": "Low",
"link": "https://learn.microsoft.com/purview/purview"
},
{
"category": "Governance and Security",
"subcategory": "Encryption at Rest",
"text": "Encrypt data used for RAG with SSE/Disk encryption with optional BYOK",
"waf": "Security",
"service": "OpenAI",
"guid": "4fda1dbf-3dd9-45d4-ac7c-891dca1f6d56",
"id": "AOAI.34",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/use-your-data-securely"
},
{
"category": "Governance and Security",
"subcategory": "Transit Encryption",
"text": "Ensure TLS is enforced for data in transit across data sources, AI search used for Retrieval-Augmented Generation (RAG) and LLM communication",
"waf": "Security",
"service": "OpenAI",
"guid": "59ae558b-937d-4498-9e11-12dbd7ba012f",
"id": "AOAI.35",
"severity": "High",
"link": "https://learn.microsoft.com/azure/search/search-security-overview"
},
{
"category": "Governance and Security",
"subcategory": "Access Control",
"text": "Use RBAC to manage access to Azure OpenAI services. Assign appropriate permissions to users and restrict access based on their roles and responsibilities",
"waf": "Security",
"service": "OpenAI",
"guid": "7b94ef6e-047d-42ea-8992-b1cd6e2054b2",
"id": "AOAI.36",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/role-based-access-control"
},
{
"category": "Governance and Security",
"subcategory": "Data Masking and Redaction",
"text": "Implement data encryption, masking or redaction techniques to hide sensitive data or replace it with obfuscated values in non-production environments or when sharing data for testing or troubleshooting purposes",
"waf": "Security",
"service": "OpenAI",
"guid": "9769e4a6-91e8-4838-ac93-6667e13c0056",
"id": "AOAI.37",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/security/fundamentals/data-encryption-best-practices"
},
{
"category": "Governance and Security",
"subcategory": "Threat Detection and Monitoring",
"text": "Utilize Azure Defender to detect and respond to security threats and set up monitoring and alerting mechanisms to identify suspicious activities or breaches. Leverage Azure Sentinel for advanced threat detection and response",
"waf": "Security",
"service": "OpenAI",
"guid": "74b1e945-b459-4837-be7a-d6c6d3b375a5",
"id": "AOAI.38",
"severity": "High",
"link": "https://learn.microsoft.com/azure/defender-for-cloud/ai-onboarding"
},
{
"category": "Governance and Security",
"subcategory": "Data Retention and Disposal",
"text": "Establish data retention and disposal policies to adhere to compliance regulations. Implement secure deletion methods for data that is no longer required and maintain an audit trail of data retention and disposal activities",
"waf": "Security",
"service": "OpenAI",
"guid": "c7acbe48-abe5-44cd-99f2-e87768468c55",
"id": "AOAI.39",
"severity": "Medium",
"link": "https://techcommunity.microsoft.com/t5/azure-storage-blog/managing-long-term-log-retention-or-any-business-data/ba-p/2494791"
},
{
"category": "Responsible AI",
"subcategory": "Jail break Safety",
"text": "Implement Prompt shields and groundedness detection using Content Safety ",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "a9c27d9c-42bb-46bd-8c69-99a246f3389a",
"id": "AOAI.4",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/content-safety/concepts/jailbreak-detection"
},
{
"category": "Governance and Security",
"subcategory": "Data Privacy and Compliance",
"text": "Ensure compliance with relevant data protection regulations, such as GDPR or HIPAA, by implementing privacy controls and obtaining necessary consents or permissions for data processing activities.",
"waf": "Security",
"service": "OpenAI",
"guid": "a775c6ee-95b9-46ad-a844-ce3b2b38b876",
"id": "AOAI.40",
"severity": "High",
"link": "https://learn.microsoft.com/azure/compliance/"
},
{
"category": "Governance and Security",
"subcategory": "Employee Awareness and Training",
"text": "Educate your employees about data security best practices, the importance of handling data securely, and potential risks associated with data breaches. Encourage them to follow data security protocols diligently.",
"waf": "Security",
"service": "OpenAI",
"guid": "ba2cf204-9901-43a5-b3ce-474dccbd8682",
"id": "AOAI.41",
"severity": "Medium"
},
{
"category": "Governance and Security",
"subcategory": "Environment segregation",
"text": "Keep production data separate from development and testing data. Only use real sensitive data in production and utilize anonymized or synthetic data in development and test environments.",
"waf": "Security",
"service": "OpenAI",
"guid": "eae01e6e-842e-452f-9721-d928c1b1cd52",
"id": "AOAI.42",
"severity": "High"
},
{
"category": "Governance and Security",
"subcategory": "Index Segregation",
"text": "If you have varying levels of data sensitivity, consider creating separate indexes for each level. For instance, you could have one index for general data and another for sensitive data, each governed by different access protocols",
"waf": "Security",
"service": "OpenAI",
"guid": "1e54a29a-9de3-499c-bd7b-28dc93555620",
"id": "AOAI.43",
"severity": "Medium"
},
{
"category": "Governance and Security",
"subcategory": "Sensitive Data in Separate Instances",
"text": "Take segregation a step further by placing sensitive datasets in different instances of the service. Each instance can be controlled with its own specific set of RBAC policies",
"waf": "Security",
"service": "OpenAI",
"guid": "2bfe4564-b0d8-434a-948b-263e6dd60512",
"id": "AOAI.44",
"severity": "Medium"
},
{
"category": "Governance and Security",
"subcategory": "Embedding and Vector handling",
"text": "Recognize that embeddings and vectors generated from sensitive information are themselves sensitive. This data should be afforded the same protective measures as the source material",
"waf": "Security",
"service": "OpenAI",
"guid": "a36498f6-dbad-438e-ad53-cc7ce1d7aaab",
"id": "AOAI.45",
"severity": "High"
},
{
"category": "Governance and Security",
"subcategory": "Access control",
"text": "Apply RBAC to th data stores having embeddings and vectors and scope access based on role's access requirements",
"waf": "Security",
"service": "OpenAI",
"guid": "3571449a-b805-43d8-af89-dc7b33be2a1a",
"id": "AOAI.46",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/role-based-access-control"
},
{
"category": "Governance and Security",
"subcategory": "Network security",
"text": "Configure private endpoint for AI services to restrict service access within your network",
"waf": "Security",
"service": "OpenAI",
"guid": "27f7b9e9-1be1-4f38-aef3-9812bd463cbb",
"id": "AOAI.47",
"graph": "resources | where type =~ 'Microsoft.CognitiveServices/accounts' or type == 'microsoft.search/searchservices' | project id, compliant = (properties.privateEndpointConnections != '[]' and properties.publicNetworkAccess !~ 'enabled')",
"severity": "High",
"link": "https://techcommunity.microsoft.com/t5/azure-architecture-blog/azure-openai-private-endpoints-connecting-across-vnet-s/ba-p/3913325"
},
{
"category": "Governance and Security",
"subcategory": "Network security",
"text": "Enforce strict inbound and outbound traffic control with Azure Firewall and UDRs and limit the external integration points",
"waf": "Security",
"service": "OpenAI",
"guid": "ac8ac199-ebb9-41a3-9d90-cae2cc881370",
"id": "AOAI.48",
"severity": "High"
},
{
"category": "Governance and Security",
"subcategory": "Control Network Access",
"text": "Implement network segmentation and access controls to restrict access to the LLM application only to authorized users and systems and prevent lateral movement",
"waf": "Security",
"service": "OpenAI",
"guid": "6f7c0cba-fe51-4464-add4-57e927138b82",
"id": "AOAI.49",
"severity": "High"
},
{
"category": "Cost Optimization",
"subcategory": "Token Optimization",
"text": "Use prompt compression tools like LLMLingua or gprtrim",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "7f42c78e-78cb-46a2-8ad1-90916e6a8d8f",
"id": "AOAI.5",
"severity": "Medium",
"link": "https://www.microsoft.com/research/blog/llmlingua-innovating-llm-efficiency-with-prompt-compression/"
},
{
"category": "Governance and Security",
"subcategory": "Secure APIs and Endpoints",
"text": "Ensure that APIs and endpoints used by the LLM application are properly secured with authentication and authorization mechanisms, such as Managed identities, API keys or OAuth, to prevent unauthorized access.",
"waf": "Security",
"service": "OpenAI",
"guid": "1102cac6-eae0-41e6-b842-e52f4721d928",
"id": "AOAI.50",
"graph": "resources | where type =~ 'Microsoft.CognitiveServices/accounts' or type == 'microsoft.search/searchservices' | project id, compliant = (isnotnull(identity))",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/managed-identity"
},
{
"category": "Governance and Security",
"subcategory": "Implement Strong Authentication",
"text": "Enforce strong end user authentication mechanisms, such as multi-factor authentication, to prevent unauthorized access to the LLM application and associated network resources",
"waf": "Security",
"service": "OpenAI",
"guid": "c1b1cd52-1e54-4a29-a9de-399cfd7b28dc",
"id": "AOAI.51",
"severity": "Medium",
"link": "https://techcommunity.microsoft.com/t5/azure-architecture-blog/security-best-practices-for-genai-applications-openai-in-azure/ba-p/4027885"
},
{
"category": "Governance and Security",
"subcategory": "Use Network Monitoring",
"text": "Implement network monitoring tools to detect and analyze network traffic for any suspicious or malicious activities. Enable logging to capture network events and facilitate forensic analysis in case of security incidents",
"waf": "Security",
"service": "OpenAI",
"guid": "93555620-2bfe-4456-9b0d-834a348b263e",
"id": "AOAI.52",
"severity": "Medium"
},
{
"category": "Governance and Security",
"subcategory": "Security Audits and Penetration Testing",
"text": "Conduct security audits and penetration testing to identify and address any network security weaknesses or vulnerabilities in the LLM application's network infrastructure",
"waf": "Security",
"service": "OpenAI",
"guid": "6dd60512-a364-498f-9dba-d38ead53cc7c",
"id": "AOAI.53",
"severity": "Medium"
},
{
"category": "Governance and Security",
"subcategory": "Infrastructure Deployment",
"text": "Azure AI Services are properly tagged for better management",
"waf": "Operational Excellence",
"guid": "e1d7aaab-3571-4449-ab80-53d89f89dc7b",
"id": "AOAI.54",
"graph": "resources | where type == 'microsoft.cognitiveservices/accounts' or type == 'microsoft.search/searchservices' | project id, compliant = (tags != '{}')",
"service": "OpenAI",
"severity": "Low",
"link": "https://learn.microsoft.com/azure/azure-resource-manager/management/tag-resources?tabs=json"
},
{
"category": "Governance and Security",
"subcategory": "Infrastructure Deployment",
"text": "Azure AI Service accounts follows organizational naming conventions",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "77036e5e-6b4b-4ed3-b503-547c1347dc56",
"id": "AOAI.55",
"severity": "Low",
"link": "https://learn.microsoft.com/azure/cloud-adoption-framework/ready/azure-best-practices/resource-abbreviations"
},
{
"category": "Governance and Security",
"subcategory": "Diagnostics Logging",
"text": "Diagnostic logs in Azure AI services resources should be enabled",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "028a71ff-e1ce-415d-b3f0-d5e772d41e36",
"id": "AOAI.56",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/diagnostic-logging"
},
{
"category": "Identity and Access Management",
"subcategory": "Entra ID based access",
"text": "Key access (local authentication) is recommended to be disabled for security. After disabling key based access, Microsoft Entra ID becomes the only access method, which allows maintaining minimum privilege principle and granular control. ",
"waf": "Security",
"service": "OpenAI",
"guid": "11cc57b4-a4b1-4410-b439-58a8c2289b3d",
"id": "AOAI.57",
"graph": "resources | where type =~ 'Microsoft.CognitiveServices/accounts' or type == 'microsoft.search/searchservices' | project id, compliant = (properties.disableLocalAuth == true)",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/authentication"
},
{
"category": "Governance and Security",
"subcategory": "Secure Key Management",
"text": "Store and manage keys securely using Azure Key Vault. Avoid hard-coding or embedding sensitive keys within your LLM application's code and retrieve them securely from Azure Key Vault using managed identities",
"waf": "Security",
"service": "OpenAI",
"guid": "6b57cfc6-5546-41e1-a3e3-453a3c863964",
"id": "AOAI.58",
"severity": "High",
"link": "https://learn.microsoft.com/azure/key-vault/general/best-practices"
},
{
"category": "Governance and Security",
"subcategory": "Key Rotation and Expiration",
"text": "Regularly rotate and expire keys stored in Azure Key Vault to minimize the risk of unauthorized access.",
"waf": "Security",
"service": "OpenAI",
"guid": "8b652d6c-15f5-4129-9539-8e6ded227dd1",
"id": "AOAI.59",
"severity": "High",
"link": "https://learn.microsoft.com/azure/key-vault/general/best-practices"
},
{
"category": "Cost Optimization",
"subcategory": "Token Optimization",
"text": "Use tiktoken to understand token sizes for token optimizations in conversational mode",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "adfe27be-e297-401a-a352-baaab79b088d",
"id": "AOAI.6",
"severity": "High",
"link": "https://github.com/openai/tiktoken"
},
{
"category": "Governance and Security",
"subcategory": "Secure coding practice",
"text": "Follow secure coding practices to prevent common vulnerabilities such as injection attacks, cross-site scripting (XSS), or security misconfigurations",
"waf": "Security",
"service": "OpenAI",
"guid": "42b06c21-d799-49a6-96f4-389a7f42c78e",
"id": "AOAI.60",
"severity": "High",
"link": "https://learn.microsoft.com/azure/security/develop/secure-dev-overview"
},
{
"category": "Governance and Security",
"subcategory": "Patching and updates",
"text": "Setup a process to regularly update and patch the LLM libraries and other system components",
"waf": "Security",
"service": "OpenAI",
"guid": "78c06a73-a22a-4495-9e6a-8dc4a20e27c3",
"id": "AOAI.61",
"severity": "High",
"link": "https://learn.microsoft.com/azure/devops/repos/security/github-advanced-security-dependency-scanning?view=azure-devops"
},
{
"category": "Responsible AI",
"subcategory": "Governance",
"text": "Adhere to Azure OpenAI or other LLMs terms of use, policies and guidance and allowed use cases",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "e29711b1-352b-4eee-879b-588defc4972c",
"id": "AOAI.62",
"severity": "High",
"link": "https://learn.microsoft.com/legal/cognitive-services/openai/code-of-conduct"
},
{
"category": "Cost Optimization",
"subcategory": "Cost familiarization",
"text": "Understand difference in cost of base models and fine tuned models and token step sizes",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "d3cd21bf-7703-46e5-b6b4-bed3d503547c",
"id": "AOAI.63",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs#base-series-and-codex-series-fine-tuned-models"
},
{
"category": "Cost Optimization",
"subcategory": "Batch processing",
"text": "Batch requests, where possible, to minimize the per-call overhead which can reduce overall costs. Ensure you optimize batch size",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "1347dc56-028a-471f-be1c-e15dd3f0d5e7",
"id": "AOAI.64",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/latency#batching"
},
{
"category": "Cost Optimization",
"subcategory": "Cost monitoring",
"text": "Set up a cost tracking system that monitors model usage and use that information to help inform model choices and prompt sizes",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "72d41e36-11cc-457b-9a4b-1410d43958a8",
"id": "AOAI.65",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs"
},
{
"category": "Cost Optimization",
"subcategory": "Token limit",
"text": "Set a maximum limit on the number of tokens per model response (max_tokens and the number of completions to generate). Optimize the size to ensure it is large enough for a valid response",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "166cd072-af9b-4141-a898-a535e737897e",
"id": "AOAI.66",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/quota?tabs=rest#understanding-rate-limits"
},
{
"category": "Operations Management",
"subcategory": "AI Search Vector Limits",
"text": "Plan and manage AI Search Vector storage",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "3266b225-86f4-4a16-92bd-ddea8a487cde",
"id": "AOAI.68",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/search/vector-search-index-size?tabs=portal-vector-quota"
},
{
"category": "Operations Management",
"subcategory": "DevOps",
"text": "Ensure deployment of Azure OpenAI instances across your various environments, such as development, test, and production supporting lrarning & experimentation. Apply LLMOps practices to automate the lifecycle management of your GenAI applications",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "b4861bc3-bc14-4aeb-9e66-e8d9a3aec218",
"id": "AOAI.69",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/machine-learning/prompt-flow/how-to-end-to-end-llmops-with-prompt-flow?view=azureml-api-2"
},
{
"category": "Cost Optimization",
"subcategory": "Costing Model",
"text": "Evaluate usage of billing models - PAYG vs PTU. Start with PAYG and consider PTU when the usage is predictable in production since it offers dedicated memory and compute, reserved capacity, and consistent maximum latency for the specified model version",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "aa80932c-8ec9-4d1b-a770-26e5e6beba9e",
"id": "AOAI.7",
"severity": "High",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/provisioned-throughput-onboarding#understanding-the-provisioned-throughput-purchase-model"
},
{
"category": "Operations Management",
"subcategory": "DevOps",
"text": "Evaluate the quality of prompts and applications when switching between model versions",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "e6436b07-36db-455f-9796-03334bdf9cc2",
"id": "AOAI.70",
"severity": "Medium",
"link": "https://techcommunity.microsoft.com/t5/ai-azure-ai-services-blog/how-to-control-azure-openai-models/ba-p/4146793"
},
{
"category": "Operations Management",
"subcategory": "Development",
"text": "Evaluate, monitor and refine your GenAI apps for features like groundedness, relevance, accuracy, coherence and fluency",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "3418db61-2712-4650-9bb4-7a393a080327",
"id": "AOAI.71",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/machine-learning/prompt-flow/concept-model-monitoring-generative-ai-evaluation-metrics?view=azureml-api-2"
},
{
"category": "Operations Management",
"subcategory": "Development",
"text": "Evaluate your Azure AI Search results based on different search parameters",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "294798b1-578b-4219-a46c-eb5443513592",
"id": "AOAI.72",
"severity": "Medium"
},
{
"category": "Operations Management",
"subcategory": "Development",
"text": "Look at fine tuning models as way of increasing accuracy only when you have tried other basic approaches like prompt engineering and RAG with your data",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "2744293b-b628-4537-a551-19b08e8f5854",
"id": "AOAI.73",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/concepts/fine-tuning-considerations"
},
{
"category": "Operations Management",
"subcategory": "Development",
"text": "Use prompt engineering techniques to improve the accuracy of LLM responses",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "287d9cec-166c-4d07-8af9-b141a898a535",
"id": "AOAI.74",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/concepts/advanced-prompt-engineering?pivots=programming-language-chat-completions"
},
{
"category": "Governance and Security",
"subcategory": "Security Audits and Penetration Testing",
"text": "Red team your GenAI applications",
"waf": "Security",
"service": "OpenAI",
"guid": "e737897e-71ca-47da-acfa-962a1594946d",
"id": "AOAI.75",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/concepts/red-teaming"
},
{
"category": "Operations Management",
"subcategory": "End user feedback",
"text": "Provide end users with scoring options for LLM responses and track these scores. ",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "edb117e6-76aa-4f66-aca4-8e5a95f2223e",
"id": "AOAI.76",
"severity": "Medium",
"link": "https://www.microsoft.com/haxtoolkit/guideline/encourage-granular-feedback/"
},
{
"category": "Cost Optimization",
"subcategory": "Quota Management",
"text": "Consider Quota management practices. Use dynamic quota for certain use cases when your application can use extra capacity opportunistically or the application itself is driving the rate at which the Azure OpenAI API is called",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "d5f3547c-c346-4d81-9028-a71ffe1b9b5d",
"id": "AOAI.8",
"severity": "High",
"link": "https://techcommunity.microsoft.com/t5/fasttrack-for-azure/optimizing-azure-openai-a-guide-to-limits-quotas-and-best/ba-p/4076268"
},
{
"category": "Operations Management",
"subcategory": "Load Balancing",
"text": "Use Load balancer solutions like APIM based gateway for balancing load and capacity across services and regions",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc410",
"id": "AOAI.9",
"severity": "Medium",
"link": "https://github.com/Azure/aoai-apim/blob/main/README.md"
},
{
"category": "Operations Management",
"subcategory": "Fine tuning",
"text": "Follow the guidance for fine-tuning with large data files and import the data from an Azure blob store. Large files, 100 MB or larger, can become unstable when uploaded through multipart forms because the requests are atomic and can't be retried or resumed",
"waf": "Reliability",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc411",
"id": "AOAI.77",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/fine-tuning?tabs=turbo%2Cpython-new&pivots=programming-language-studio#import-training-data-from-azure-blob-store"
},
{
"category": "Operations Management",
"subcategory": "Monitoring",
"text": "Manage rate limits for your model deployments and monitor usage of tokens per minute (TPM) and requests per minute (RPM) for pay-as-you-go deployments",
"waf": "Reliability",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc412",
"id": "AOAI.78",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/quota?tabs=rest"
},
{
"category": "Operations Management",
"subcategory": "Monitoring",
"text": "Monitor provision-managed utilization if you're using the provisioned throughput payment model",
"waf": "Reliability",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc413",
"id": "AOAI.79",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/monitor-openai"
},
{
"category": "Responsible AI",
"subcategory": "Content Safety",
"text": "Tune content filters to minimize false positives from overly aggressive filters",
"waf": "Reliability",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc414",
"id": "AOAI.80",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/content-filters"
},
{
"category": "Governance and Security",
"subcategory": "Key Management",
"text": "Use customer-managed keys for fine-tuned models and training data that's uploaded to Azure OpenAI",
"waf": "Security",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc415",
"id": "AOAI.81",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/encrypt-data-at-rest"
},
{
"category": "Governance and Security",
"subcategory": "Jailbreak protection",
"text": "Implement jailbreak risk detection to safeguard your language model deployments against prompt injection attacks",
"waf": "Security",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc416",
"id": "AOAI.82",
"graph": "resources | where type == 'microsoft.cognitiveservices/accounts' and kind =~ 'contentsafety' | project id, compliant = 1",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/content-safety/concepts/jailbreak-detection"
},
{
"category": "Governance and Security",
"subcategory": "Quota exhaustion",
"text": "Use security controls like throttling, service isolation and gateway pattern to prevent attacks that might exhaust model usage quotas",
"waf": "Security",
"service": "OpenAI",
"guid": "9de0d5d7-31d4-41e3-911c-817bfafbc417",
"id": "AOAI.83",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/monitor-openai"
},
{
"category": "Cost Optimization",
"subcategory": "Cost estimation",
"text": "Develop your cost model, considering prompt sizes. Understanding prompt input and response sizes and how text translates into tokens helps you create a viable cost model",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "72d41e36-11cc-457b-9a4b-1410d43958a9",
"id": "AOAI.84",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs"
},
{
"category": "Cost Optimization",
"subcategory": "Model selection",
"text": "Consider model pricing and capabilities when you choose models. Start with less-costly models for less-complex tasks like text generation or completion tasks and for complex tasks like language translation or content understanding, consider using more advanced models. Optimize costs while still achieving the desired application performance",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "72d41e36-11cc-457b-9a4b-1410d43958a1",
"id": "AOAI.85",
"severity": "Medium",
"link": "https://azure.microsoft.com/pricing/details/cognitive-services/openai-service/"
},
{
"category": "Cost Optimization",
"subcategory": "Usage Optimization",
"text": "Maximize Azure OpenAI price breakpoints like fine-tuning and model breakpoints like image generation to your advantage. Fine-tuning is charged per hour, use as much time as you have available per hour to improve results without slipping into the next billing period. The cost for generating 100 images is the same as the cost for 1 image",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "72d41e36-11cc-457b-9a4b-1410d43958a2",
"id": "AOAI.86",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs"
},
{
"category": "Cost Optimization",
"subcategory": "Usage Optimization",
"text": "Remove unused fine-tuned models when they're no longer being consumed to avoid incurring an ongoing hosting fee",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "72d41e36-11cc-457b-9a4b-1410d43958a3",
"id": "AOAI.87",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs"
},
{
"category": "Cost Optimization",
"subcategory": "Token Optimization",
"text": "Create concise prompts that provide enough context for the model to generate a useful response. Also ensure that you optimize the limit of the response length.",
"waf": "Cost Optimization",
"service": "OpenAI",
"guid": "7f42c78e-78cb-46a2-8ad1-90916e6a8d8g",
"id": "AOAI.88",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/openai/how-to/manage-costs"
},
{
"category": "Operations Management",
"subcategory": "IaC",
"text": "Use infrastructure as code (IaC) to deploy Azure OpenAI, model deployments, and other infrastructure required for fine-tuning models",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "b4861bc3-bc14-4aeb-9e66-e8d9a3aec219",
"id": "AOAI.89",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/ai-services/create-account-bicep"
},
{
"category": "Operations Management",
"subcategory": "Development",
"text": "Consider using dedicated model deployments per consumer group to provide per-model usage isolation that can help prevent noisy neighbors between your consumer groups",
"waf": "Operational Excellence",
"service": "OpenAI",
"guid": "2744293b-b628-4537-a551-19b08e8f5855",
"id": "AOAI.90",
"severity": "Medium",
"link": "https://learn.microsoft.com/azure/architecture/guide/multitenant/service/openai"
}
],
"categories": [
{
"name": "Identity and Access Management"
},
{
"name": "Network Topology and Connectivity"
},
{
"name": "BC and DR"
},
{
"name": "Governance and Security"
},
{
"name": "Cost Governance"
},
{
"name": "Operations Management"
},
{
"name": "Application Deployment"