Skip to content

Latest commit

 

History

History
88 lines (58 loc) · 4.29 KB

File metadata and controls

88 lines (58 loc) · 4.29 KB

AI Act Compliance Agents

An agent-harness concept for structured, fictional or fully sanitized AI traceability drafts and review artifacts.

Maturity

Early practitioner prototype.

This repository is a structured drafting and traceability aid. It is not a legal product, compliance certification tool, safety-case generator, or substitute for qualified legal, safety, cybersecurity, privacy, regulatory, or engineering review.

Start here

Purpose

The repository helps product, safety, compliance, and engineering teams explore repeatable structures for:

  • EU AI Act risk-classification support artifacts
  • technical-documentation drafts
  • human-oversight design patterns
  • post-market-monitoring plan drafts
  • traceability between stated controls, source references, ownership, review state, and evidence pointers

Current capabilities

  • specialized agent, skill, command, template, and rule materials for practitioner workflows
  • lightweight Python utility for structured traceability rows
  • schemas/traceability.schema.json machine-readable input schema
  • validation for stable control IDs, source framework/reference/version, owner, review status, and evidence records
  • fictional sample input and unit tests
  • an explicit input contract, methodology, source baseline, and release-preflight checklist

Quick start

  1. Install in editable mode: python -m pip install -e .
  2. Review the traceability input contract.
  3. Run the fictional example: ai-act-trace examples/sample-requirements.json
  4. Read METHODOLOGY.md and SOURCES.md before adapting any structure.

Traceability input migration

The utility expects structured control objects, not a list of plain strings. Each control contains a stable ID, source framework/reference/version, accountable owner, review status, and evidence list.

This supports traceability and review preparation; it does not make a control legally sufficient or technically verified.

Publication safety

Only fictional, synthetic, or fully sanitized material belongs here.

Do not publish:

  • real company compliance positions or internal risk classifications
  • confidential legal interpretations, safety analyses, or security/privacy assessments
  • proprietary feature behavior, release gates, architecture, roadmaps, supplier details, or test results
  • real safety requirements, hazards, incidents, evidence, or conformity conclusions
  • licensed standards text, employer policies, credentials, endpoints, or personal data

Out of scope

This prototype does not provide:

  • legal advice, compliance certification, or EU AI Act conformity assessment
  • ISO 26262, ISO 21448, or safety-case certification
  • regulatory or production-release approval
  • evidence-quality validation, legal applicability determination, or authoritative standards interpretation
  • official guidance from any standards body, regulator, employer, or automaker

Next quality steps

  1. broader edge-case coverage for nested evidence, lifecycle updates, and reviewer handover
  2. export formats that preserve source provenance and review history
  3. a controlled example of lifecycle status changes using only fictional information
  4. a documented source-refresh process for maintained public examples

Scope and disclaimer

This repository is shared in a personal capacity. It is not affiliated with or endorsed by any automaker, supplier, regulator, standards body, or employer.

References to laws, standards, conformity assessment, human oversight, post-market monitoring, or safety alignment are practitioner examples for structured thinking. Verify against official texts and use qualified review before applying any artifact to a real system, compliance process, safety process, or release decision.


Maintained by Sima Bagheri.