Model Security Operations¶
This guide walks through managing AI Model Security using the daystrom model-security command group — security groups, rules, rule instances, scans, and labels.
All output shown below is captured from real Prisma AIRS API responses.
Security Groups¶
List groups¶
Prisma AIRS — Model Security
ML model supply chain security
Security Groups:
bb1d038a-0506-4b07-8f16-a723b8c1a1c7
Default GCS ACTIVE source: GCS
020d546d-3920-4ef3-9183-00f37f33f566
Default LOCAL ACTIVE source: LOCAL
6a1e67e1-00cc-45dc-9395-3a9e3dbf50f9
Default S3 ACTIVE source: S3
fd1a4209-32d0-4a1a-bd40-cde35104dc39
Default AZURE ACTIVE source: AZURE
4c22aef7-2ab7-40ba-b3f1-cd9e9aa1768e
Default HUGGING_FACE ACTIVE source: HUGGING_FACE
Filter groups by source type¶
Security Groups:
020d546d-3920-4ef3-9183-00f37f33f566
Default LOCAL ACTIVE source: LOCAL
6a1e67e1-00cc-45dc-9395-3a9e3dbf50f9
Default S3 ACTIVE source: S3
Get group details¶
Security Group Detail:
UUID: bb1d038a-0506-4b07-8f16-a723b8c1a1c7
Name: Default GCS
Description: Auto-created default security group for GCS models
Source Type: GCS
State: ACTIVE
Created: 2025-10-30T14:14:17.321066Z
Updated: 2026-02-19T16:03:26.660748Z
Create a group¶
{
"name": "Custom S3 Group",
"source_type": "S3",
"description": "Custom security group for S3 model sources"
}
Update and delete¶
daystrom model-security groups update <uuid> --name "Renamed Group"
daystrom model-security groups delete <uuid>
Security Rules¶
Rules define the security checks applied to models. They are read-only — managed by Prisma AIRS.
List rules¶
Security Rules:
550e8400-e29b-41d4-a716-44665544000b
Known Framework Operators Check type: ARTIFACT default: BLOCKING
Model artifacts should only contain known safe TensorFlow operators
Sources: ALL
550e8400-e29b-41d4-a716-446655440006
License Exists type: METADATA default: BLOCKING
Models should have a license
Sources: HUGGING_FACE
550e8400-e29b-41d4-a716-446655440007
License Is Valid For Use type: METADATA default: BLOCKING
Models license should be valid for use
Sources: HUGGING_FACE
550e8400-e29b-41d4-a716-446655440008
Load Time Code Execution Check type: ARTIFACT default: BLOCKING
Model artifacts should not contain unsafe operators that are run upon deserialization
Sources: ALL
550e8400-e29b-41d4-a716-44665544000a
Model Architecture Backdoor Check type: ARTIFACT default: BLOCKING
Model architecture files should not contain backdoor attacks
Sources: ALL
Search rules¶
Security Rules:
550e8400-e29b-41d4-a716-446655440006
License Exists type: METADATA default: BLOCKING
Models should have a license
Sources: HUGGING_FACE
550e8400-e29b-41d4-a716-446655440007
License Is Valid For Use type: METADATA default: BLOCKING
Models license should be valid for use
Sources: HUGGING_FACE
Get rule details¶
Security Rule Detail:
UUID: 550e8400-e29b-41d4-a716-44665544000b
Name: Known Framework Operators Check
Description: Model artifacts should only contain known safe TensorFlow operators
Rule Type: ARTIFACT
Default State: BLOCKING
Sources: ALL
Remediation:
Ensure that the model does not contain any custom TensorFlow operators
https://docs.paloaltonetworks.com/.../known-framework-operators-check
Rule Instances¶
Rule instances are the per-group configuration of security rules.
List rule instances¶
Rule Instances:
67185c67-020a-4625-816f-9a2137e3d6b3
Known Framework Operators Check BLOCKING
34f0a130-d2a3-451d-8795-776f5d5bdd8f
Load Time Code Execution Check BLOCKING
0fe19730-3fb4-474b-9160-d2de02be592b
Model Architecture Backdoor Check BLOCKING
c532d04b-ca2d-4e03-8b4b-8a81917c02cf
Runtime Code Execution Check BLOCKING
53aab919-06c4-4ebf-a1e1-8124ce2ae0d5
Stored In Approved File Format BLOCKING
6648097a-37bd-4726-ab51-a63188c52a23
Stored In Approved Location DISABLED
43dd1d74-f902-4337-8615-85c7e7403098
Suspicious Model Components Check BLOCKING
Filter by state¶
Rule Instances:
67185c67-020a-4625-816f-9a2137e3d6b3
Known Framework Operators Check BLOCKING
34f0a130-d2a3-451d-8795-776f5d5bdd8f
Load Time Code Execution Check BLOCKING
0fe19730-3fb4-474b-9160-d2de02be592b
Model Architecture Backdoor Check BLOCKING
c532d04b-ca2d-4e03-8b4b-8a81917c02cf
Runtime Code Execution Check BLOCKING
53aab919-06c4-4ebf-a1e1-8124ce2ae0d5
Stored In Approved File Format BLOCKING
43dd1d74-f902-4337-8615-85c7e7403098
Suspicious Model Components Check BLOCKING
Get rule instance details¶
daystrom model-security rule-instances get 020d546d-3920-4ef3-9183-00f37f33f566 67185c67-020a-4625-816f-9a2137e3d6b3
Rule Instance Detail:
UUID: 67185c67-020a-4625-816f-9a2137e3d6b3
Group UUID: 020d546d-3920-4ef3-9183-00f37f33f566
Rule UUID: 550e8400-e29b-41d4-a716-44665544000b
State: BLOCKING
Rule Name: Known Framework Operators Check
Created: 2025-10-30T14:14:17.321066Z
Updated: 2025-10-30T14:14:17.321066Z
Update a rule instance¶
daystrom model-security rule-instances update <groupUuid> <instanceUuid> --config rule-instance-update.json
Scans¶
List scans¶
Model Security Scans:
7a7e1cdf-a6b1-4743-a5f2-a7bd96ec7bab
BLOCKED HUGGING_FACE 2026-03-03T22:32:12.344402Z
https://huggingface.co/microsoft/DialoGPT-medium
Rules: 10 passed 1 failed / 11 total
ee71b4da-64ce-4d6c-96fb-2bced1154a06
ALLOWED MODEL_SECURITY_SDK 2026-03-03T22:21:44.130386Z
/Users/.../models/qwen3-0.6b-saffron-merged
Rules: 6 passed 0 failed / 6 total
6456c1fd-1a76-40c4-a591-3275d7885c17
ALLOWED MODEL_SECURITY_SDK 2026-03-03T22:16:54.578468Z
models/qwen3-0.6b-saffron-qlora/
Rules: 6 passed 0 failed / 6 total
Filter scans¶
# By evaluation outcome
daystrom model-security scans list --eval-outcome BLOCKED
# By source type
daystrom model-security scans list --source-type HUGGING_FACE
Get scan details¶
Scan Detail:
UUID: 7a7e1cdf-a6b1-4743-a5f2-a7bd96ec7bab
Outcome: BLOCKED
Model URI: https://huggingface.co/microsoft/DialoGPT-medium
Origin: HUGGING_FACE
Source: HUGGING_FACE
Group: Default HUGGING_FACE
Created: 2026-03-03T22:32:12.344402Z
Updated: 2026-03-03T22:32:12.477430Z
Rules: 10 passed 1 failed / 11 total
View scan evaluations¶
Rule Evaluations:
83924890-6823-4e6f-9541-3eed300b890e
License Exists PASSED BLOCKING
1a4db8ab-cbe6-4f9f-bfbf-fb65fdb451de
License Is Valid For Use PASSED BLOCKING
0e424aa6-0fb7-427a-8809-3caa301a8c40
Model Is Blocked PASSED BLOCKING
ec06ca86-1271-4d3a-a310-272190f74503
Organization Is Blocked PASSED BLOCKING
494c926e-0f8a-4a22-87ae-2a1ff06c0e7b
Organization Verified By Hugging Face PASSED BLOCKING
1ae405a2-15a9-4a3b-8e57-5a916c133857
Stored In Approved File Format FAILED BLOCKING
6c5b370e-320b-41bb-b206-8a4e5265eeea
Known Framework Operators Check PASSED BLOCKING
779f12dd-b672-494f-9962-2e642ff3716e
Load Time Code Execution Check PASSED BLOCKING
5fbe2ef0-6cb8-441d-b6cf-88ff1c6f9a93
Model Architecture Backdoor Check PASSED BLOCKING
60c3cd09-e9ce-40f2-aff3-d62c21c9c64f
Runtime Code Execution Check PASSED BLOCKING
33fc876c-18b1-4c58-863e-c133fdb07fb1
Suspicious Model Components Check PASSED BLOCKING
View violations¶
Violations:
0115bae0-bc07-468b-8ca6-a585d8c82f59
Stored In Approved File Format pytorch_model.bin
Model file 'pytorch_model.bin' is stored in an unapproved format: pickle
Threat: UNAPPROVED_FORMATS
2b5e4d7b-2a50-408e-b6cf-640f1b8ec4cd
Stored In Approved File Format pytorch_model.bin
Model file 'pytorch_model.bin' is stored in an unapproved format: pytorch_v0_1_10
Threat: UNAPPROVED_FORMATS
33b437c4-47f0-4878-a249-64be8bd46e06
Stored In Approved File Format rust_model.ot
Model file 'rust_model.ot' is stored in an unapproved format: pytorch_torch_script
Threat: UNAPPROVED_FORMATS
f1409208-b28c-4234-b015-137b6b28508c
Stored In Approved File Format rust_model.ot
Model file 'rust_model.ot' is stored in an unapproved format: zip
Threat: UNAPPROVED_FORMATS
9d39c991-e37f-4e6f-bc27-bffed8d90047
Stored In Approved File Format tf_model.h5
Model file 'tf_model.h5' is stored in an unapproved format: keras_weights
Threat: UNAPPROVED_FORMATS
View scanned files¶
Scanned Files:
SUCCESS FILE config.json [json]
SKIPPED FILE flax_model.msgpack
SUCCESS FILE generation_config_for_conversational.json [json]
SUCCESS FILE generation_config.json [json]
SKIPPED FILE .gitattributes
SKIPPED FILE merges.txt
FAILED FILE pytorch_model.bin [pickle, pytorch_v0_1_10]
SKIPPED FILE README.md [not_model]
FAILED FILE rust_model.ot [pytorch_torch_script, zip]
FAILED FILE tf_model.h5 [keras_weights]
SUCCESS FILE tokenizer_config.json [json]
SUCCESS FILE vocab.json [json]
Labels¶
Labels help organize and categorize scans with key-value metadata.
Browse label taxonomy¶
Add labels¶
Set labels (replace all)¶
daystrom model-security labels set <scanUuid> --labels '[{"key":"env","value":"staging"},{"key":"team","value":"ml"}]'
Delete labels¶
PyPI Authentication¶
Get authentication URL for Google Artifact Registry (used for model scanning tools).
Common Workflows¶
Investigate a blocked scan¶
-
Find blocked scans:
-
View evaluations to find which rule failed:
-
View specific violations:
-
Check the scanned files:
-
Look up remediation steps for the failed rule:
Customize rule enforcement¶
-
List rule instances in a group:
-
Change a rule from ALLOWING to BLOCKING: