EU AI Act

The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) introduces tiered obligations for AI systems based on risk. The platform supports compliance through tooling, documentation, and controls that map to the Act's requirements.

How the platform helps

The in-platform Compliance Wizard classifies a workflow under the Act and produces an audit-ready PDF.
Workflow versioning and the requirements checklist support evidence generation for the obligations that apply at each risk tier.
Audit-grade traces supply the per-execution evidence needed for high-risk systems.

Risk tiers and obligations

Risk tier	What it covers	Platform support
Prohibited	Banned uses (e.g. social scoring, manipulative techniques).	The wizard surfaces clearly when a workflow falls in this tier so you do not deploy it.
High risk	Use in HR, credit scoring, critical infrastructure, etc.	Substantial: audit trail, versioning, requirements checklist, human-in-the-loop guidance.
Limited risk	Chatbots, content generators.	Transparency support: workflow descriptions, watermarking guidance.
Minimal risk	Common AI features (spam filters, recommenders).	Documentation only.

Per-tier obligations and platform controls

Prohibited

The platform does not block prohibited uses technically - that remains the deployer's decision. The wizard makes the classification explicit so the decision is informed.

High risk

Article 9 onwards introduces obligations for high-risk AI systems. Mapping to platform features:

Risk management system (Art. 9) - the wizard captures the risk assessment; combine with your existing risk-management process.
Data governance (Art. 10) - training data is the model vendor's concern; for fine-tuning or retrieval you control via the knowledge base. Document data provenance per collection.
Technical documentation (Art. 11) - generate the workflow description PDF and the compliance assessment PDF; these supply the technical documentation source material.
Record-keeping (Art. 12) - the platform's run history and audit trail provide automatic logging.
Transparency and information (Art. 13) - surface model name and workflow purpose to end users (deployer responsibility).
Human oversight (Art. 14) - design workflows with explicit human-in-the-loop steps for high-risk decisions; the platform supports approval-gate node patterns.
Accuracy, robustness, cybersecurity (Art. 15) - validated through ongoing evaluation. Hardening is administrator-driven.

Limited risk

Article 50 introduces transparency obligations for AI systems interacting with humans:

Disclosure - your interface must inform users they are interacting with AI. The platform does not auto-add disclosures; build them into your workflow's user-facing messages.
Watermarking of AI-generated content - for systems that generate synthetic media, follow Article 50 watermarking obligations. Tooling support depends on your media type and is outside platform scope.

Minimal risk

No Act obligations beyond general ones. Document the workflow's purpose and model selection for internal traceability.

Workflow change management under the Act

When a workflow changes substantively (different model, different purpose, expanded data scope), re-run the wizard. Because:

The risk classification might change.
The requirements checklist might expand.
The audit trail must reflect the change.

The platform versions every workflow change, so the audit trail naturally tracks substantive change events.

What the platform does not do

It does not perform conformity assessment for you. A high-risk system requires a notified-body assessment in many cases. The wizard supplies inputs; you do the assessment.
It does not classify use case automatically. The wizard presents questions; you answer them based on the deployer's knowledge of the use case.
It does not generate Article 6 high-risk Annex III evidence automatically. The PDF supplies structured documentation; the evidence work is the deployer's.

Resources

The wizard PDF you generate per workflow.
The platform's Compliance Wizard feature page.
The official EU AI Act text and the Commission's guidance, kept up to date by your legal team.

How the platform helps

The in-platform Compliance Wizard classifies a workflow under the Act and produces an audit-ready PDF.

Workflow versioning and the requirements checklist support evidence generation for the obligations that apply at each risk tier.

Audit-grade traces supply the per-execution evidence needed for high-risk systems.

Risk tiers and obligations

Risk tier	What it covers	Platform support
Prohibited	Banned uses (e.g. social scoring, manipulative techniques).	The wizard surfaces clearly when a workflow falls in this tier so you do not deploy it.
High risk	Use in HR, credit scoring, critical infrastructure, etc.	Substantial: audit trail, versioning, requirements checklist, human-in-the-loop guidance.
Limited risk	Chatbots, content generators.	Transparency support: workflow descriptions, watermarking guidance.
Minimal risk	Common AI features (spam filters, recommenders).	Documentation only.

Per-tier obligations and platform controls

The platform does not block prohibited uses technically - that remains the deployer's decision. The wizard makes the classification explicit so the decision is informed.

Article 9 onwards introduces obligations for high-risk AI systems. Mapping to platform features:

Risk management system (Art. 9) - the wizard captures the risk assessment; combine with your existing risk-management process.

Data governance (Art. 10) - training data is the model vendor's concern; for fine-tuning or retrieval you control via the knowledge base. Document data provenance per collection.

Technical documentation (Art. 11) - generate the workflow description PDF and the compliance assessment PDF; these supply the technical documentation source material.

Record-keeping (Art. 12) - the platform's run history and audit trail provide automatic logging.

Transparency and information (Art. 13) - surface model name and workflow purpose to end users (deployer responsibility).

Human oversight (Art. 14) - design workflows with explicit human-in-the-loop steps for high-risk decisions; the platform supports approval-gate node patterns.

Accuracy, robustness, cybersecurity (Art. 15) - validated through ongoing evaluation. Hardening is administrator-driven.

Article 50 introduces transparency obligations for AI systems interacting with humans:

Disclosure - your interface must inform users they are interacting with AI. The platform does not auto-add disclosures; build them into your workflow's user-facing messages.

Watermarking of AI-generated content - for systems that generate synthetic media, follow Article 50 watermarking obligations. Tooling support depends on your media type and is outside platform scope.

No Act obligations beyond general ones. Document the workflow's purpose and model selection for internal traceability.

Workflow change management under the Act

When a workflow changes substantively (different model, different purpose, expanded data scope), re-run the wizard. Because:

The risk classification might change.

The requirements checklist might expand.

The audit trail must reflect the change.

The platform versions every workflow change, so the audit trail naturally tracks substantive change events.

What the platform does not do

It does not perform conformity assessment for you. A high-risk system requires a notified-body assessment in many cases. The wizard supplies inputs; you do the assessment.

It does not classify use case automatically. The wizard presents questions; you answer them based on the deployer's knowledge of the use case.

It does not generate Article 6 high-risk Annex III evidence automatically. The PDF supplies structured documentation; the evidence work is the deployer's.