Custom System Descriptors¶

Create a YAML descriptor that conforms to AISystemDescriptor.

Required Top-Level Fields¶

name (string)
version (string)
description (string, minimum length 10)
use_cases (list, at least 1)
data_practices (list, at least 1)
human_oversight (object)
training_data_source (string, minimum length 10)
documentation (boolean)
performance_monitoring (boolean)

Optional:

incident_procedure
created_at
last_updated

Minimal Valid Template¶

name: "My AI System"
version: "1.0.0"
description: "Model that supports internal operational decision support."

use_cases:
  - domain: general_purpose
    description: "Assists support agents with response drafting."
    autonomous_decision: false
    impacts_fundamental_rights: false

data_practices:
  - type: personal
    retention_period: 30
    sharing_third_parties: false
    explicit_consent: true

human_oversight:
  oversight_mechanism: "approval_required"
  fallback_procedure: "Escalate to supervisor if confidence is low."
  review_frequency: "per_decision"
  human_authority: true

training_data_source: "Curated internal support archive with review filters."
documentation: true
performance_monitoring: true

Domain Values¶

use_cases[].domain supported enum values:

biometric
critical_infrastructure
law_enforcement
employment
credit_scoring
education
general_purpose
healthcare
content_moderation
other

Validation Workflow¶

ai-act validate my_system.yaml
ai-act classify my_system.yaml
ai-act check my_system.yaml --json

Modeling Guidance¶

Prefer concrete descriptions over generic one-liners
Document oversight and fallback procedures clearly
Keep training_data_source specific and evidence-oriented
Set documentation and performance_monitoring to reflect real controls