​

EU AI Act (Regulation 2024/1689)

​

Application timeline

​

High-risk classification

1. Article 6(1) — AI systems intended to be used as a safety component of a product covered by the Union harmonisation legislation listed in Annex I (incl. EU MR 2023/1230). Includes safety-related AI in machinery.
2. Article 6(2) — AI systems falling under the use cases listed in Annex III (biometric ID, critical infrastructure, education, employment, etc.). Most robotics applications are Article 6(1), not Article 6(2).

• Risk management system (Article 9).
• Data and data governance (Article 10) — including dataset documentation.
• Technical documentation (Article 11, Annex IV) — substantial overlap with the Machinery Regulation technical file.
• Record-keeping (Article 12).
• Transparency and provision of information to deployers (Article 13).
• Human oversight (Article 14).
• Accuracy, robustness, and cybersecurity (Article 15).

​

What Roboticks supports today

• Clause-level derivation from the AI Act, particularly Articles 9 (risk management), 12 (logging), 14 (human oversight), 15 (accuracy/robustness/cybersecurity).
• Cross-derivation with EU MR 2023/1230 — a requirement covering AI behaviour in machinery can derive from both Regulations simultaneously.
• Test evidence for AI behaviour — reproducible sim scenarios, MCAP capture of model inputs and outputs, deterministic regression tests across AI model versions.
• Accuracy/robustness evidence — JUnit aggregation of AI evaluation results; the same evidence-pack workflow applies.
• Logging evidence — Article 12 requires automatic logging; the platform’s MCAP capture of model inferences satisfies the logging requirement when the test set covers the operational scope.
• Cybersecurity evidence — SBOM, SARIF, vulnerability scanning — applies equally to AI-containing safety components.

​

What is out of scope today (Year 2 roadmap)

• Dataset cards — Article 10 expects structured documentation of training, validation, and test datasets. Roboticks does not currently author dataset cards. On the roadmap; integration with HuggingFace datasets metadata is the planned approach.
• Bias evaluations — Article 10 expects examination of training data for biases. Out of scope today; expected via connector to third-party bias-evaluation tools (e.g., Fairlearn, Aequitas) on the roadmap.
• Model cards — structured documentation of model purpose, performance, limitations. Roboticks can attach model cards to requirements as supplementary artefacts; native authoring is roadmap.
• Notified-body interaction for AI — we do not act as a notified body for AI Act conformity assessment, the same as for the Machinery Regulation.

​

Example AI-component requirement

- id: REQ-AI-007
  title: Perception model robustness to adversarial lighting
  type: safety
  asil_pl: PLd
  derives_from:
    - standard: eu-ai-act-2024-1689
      clause: "Article 15 Accuracy, robustness, and cybersecurity"
      edition: "2024-08-01"
    - standard: eu-mr-2023-1230
      clause: "Annex III, EHSR 1.1.6 AI components"
      edition: "2023-06-14"
  text: |
    The cobot perception model shall detect operator presence in
    the workspace under the lighting variation set defined in
    `scenarios/lighting_robustness.yaml`, with no false-negative
    rate exceeding 0.001% across the full evaluation set.
    The evaluation set covers nominal, low-light, high-glare,
    backlit, and adversarially-illuminated conditions.
  acceptance:
    - test: tests/ai/test_perception_robustness.py::test_no_false_negatives_under_lighting_variation

​

Suggested test patterns

​

Interaction with EU MR 2023/1230

• Conformity assessment under the Machinery Regulation may incorporate AI Act conformity assessment, depending on the route chosen and the notified body’s scope.
• Technical file content overlaps substantially (Annex IV of the Machinery Regulation vs Annex IV of the AI Act).
• Substantial-modification triggers under both Regulations apply; a change that modifies the AI model in a safety-relevant way is a substantial modification under both.

​

Pinning

rbtk standard pin eu-ai-act-2024-1689 --project acme-cobot/firmware

​

Next steps