2026-06-02·5 min read·sota.io Team

EU AI Act AI Supply Chain Compliance Finale: Complete Provider & Deployer Checklist for August 2026

Post #1456 in the sota.io EU Regulatory Compliance Series — EU AI Act AI Supply Chain 2026 #5/5 (Serie Komplett)

EU AI Act AI Supply Chain Compliance Checklist for Providers and Deployers 2026

August 2, 2026 is now weeks away. If your SaaS product integrates third-party AI APIs — OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, Google Vertex AI, Mistral, Cohere, or any other model provider — the EU AI Act's supply chain obligations apply to you right now.

This is the finale of a five-part series. We've covered who qualifies as a deployer, Art.13 and Art.26 due diligence requirements, vendor contract requirements, and Art.73 incident response obligations. Now it's time to put it all together.

This guide is the definitive checklist: 60 items across six domains, mapped to specific EU AI Act articles, organized by who owns each obligation.


Why Supply Chain Compliance Is Different

Most EU AI Act commentary focuses on AI system providers — the companies that train and release AI models. But the regulation reaches further. Under Art.3, anyone who deploys a high-risk or GPAI-enabled AI system under their own brand or integrates it into a product serving EU users is a deployer with binding obligations.

That means: if you build a SaaS product using an AI API, you are in scope. The provider (OpenAI, Anthropic, etc.) cannot absorb all your compliance obligations through their terms of service. You retain independent duties under Art.26, Art.50, and Art.73.

This is the supply chain risk that caught many SaaS teams off guard: they assumed vendor compliance cascades down automatically. It doesn't.


The Six Compliance Domains

Domain 1: Role Classification (Art.3, Art.6, Art.16, Art.26)

Before you can comply, you must classify your role in the AI value chain.

Checklist ItemArticleOwner
Determine whether your AI integration constitutes a high-risk AI system under Annex IIIArt.6Deployer
Determine whether you are acting as a provider (own-brand model or fine-tuned model)Art.3(3)Deployer
Determine whether you are acting as a deployer (integrate existing model under provider's classification)Art.3(4)Deployer
If deployer-that-modifies: re-assess whether modification triggers provider obligationsArt.25(1)Deployer
Document role classification decision with legal reasoningArt.11Deployer
Obtain confirmed role classification from AI vendor in writingArt.13Provider + Deployer

Key insight: If you fine-tune a model, instruct it with system prompts that materially change its risk profile, or integrate it into a use case not covered by the provider's conformity assessment, you may be reclassified as a provider — triggering the full Art.16 obligation set.


Domain 2: Pre-Integration Due Diligence (Art.13, Art.26)

Under Art.13, providers of high-risk AI systems must give deployers "sufficient information" to implement the system correctly. Under Art.26, deployers must verify that this information exists before integrating.

Checklist ItemArticleOwner
Request and obtain model card or system documentation from AI vendorArt.13(1)Provider → Deployer
Verify technical documentation covers: training data sources, known limitations, performance boundariesArt.13(1)(a–d)Deployer
Confirm vendor provides instructions for human oversight implementationArt.13(3)(f)Provider
Document AI system's intended purpose as specified by providerArt.13(1)(b)Provider → Deployer
Verify conformity assessment status (CE marking for high-risk systems)Art.43, Art.47Deployer
Check EU AI system database registration (Art.49 database) for high-risk systemsArt.49Deployer
Assess whether vendor documentation is sufficient for your deployment contextArt.26(2)Deployer
Document gaps between vendor documentation and your deployment requirementsArt.11Deployer

Domain 3: Contractual Requirements (Art.13, Art.25, Art.26)

Supply chain compliance cannot be achieved with standard vendor ToS. The EU AI Act requires specific provisions in contracts between AI system providers and deployers.

Checklist ItemArticleOwner
Contract specifies vendor's role as provider and your role as deployerArt.25Both
Vendor commits to provide updated technical documentation on material changesArt.13(2)Provider
Contract includes data processing and log retention obligationsArt.12, Art.26(5)Both
Vendor commits to notify deployer of serious incidents within 2 days of detectionArt.73(2)Provider
Vendor commits to cooperation with national competent authority investigationsArt.21, Art.74Provider
Contract specifies human oversight measures vendor requires deployer to implementArt.14Provider
Vendor provides EU-accessible point of contact for compliance queriesArt.16(j)Provider
Contract includes terms for audit rights if vendor is critical AI infrastructureArt.26(6)Both
SLA covers minimum uptime requirements tied to critical function continuityArt.15Provider
Contract addresses liability allocation for AI system failures under PLD 2024/2853ExternalBoth

A contract without these provisions does not meet the Art.25 standard. If your current AI vendor ToS lacks these terms, you need either amended DPAs/vendor agreements or a vendor who supports them — such as EU-native AI infrastructure providers operating under explicit EU AI Act compliance frameworks.


Domain 4: Deployment Safeguards (Art.14, Art.26, Art.50)

Once integration is complete, deployers must implement ongoing operational safeguards.

Checklist ItemArticleOwner
Implement human oversight mechanisms for high-risk AI outputsArt.14(1)Deployer
Implement kill-switch or manual override capabilityArt.14(4)(a)Deployer
Train staff responsible for AI system oversight per Art.26(6)Art.26(6)Deployer
Limit AI system use to its documented intended purposeArt.26(1)Deployer
Do not use AI system in ways that discriminate against protected groupsArt.26(3)Deployer
Implement appropriate technical and organizational measuresArt.26(2)Deployer
For GPAI/chatbot interactions: disclose AI nature to users under Art.50Art.50(1)Deployer
For AI-generated content: implement C2PA labelling or equivalentArt.50(2)Deployer
Post a public accessibility statement if EU accessibility-covered services are involvedArt.50 / EAADeployer
Maintain logs of AI system usage as technically feasibleArt.26(5)Deployer

Art.14 Human Oversight is often the most underestimated requirement. It does not mean a human reviews every AI output — it means the system is designed so a human can effectively intervene when needed. For SaaS products, this typically means: output confidence thresholds, audit trails, and administrator override controls.


Domain 5: Record-Keeping and Documentation (Art.11, Art.12, Art.17, Art.26)

The EU AI Act requires deployers to maintain records that demonstrate compliance. There is no prescribed format, but regulators expect to be able to audit the following:

Checklist ItemArticleOwner
Maintain record of AI systems integrated under deployer roleArt.26(5)Deployer
Retain vendor documentation for 10 years after last useArt.18Deployer
Maintain logs sufficient to detect post-deployment incidentsArt.19Deployer
Document fundamental rights impact assessment for high-risk AI deploymentsArt.26(9)Deployer (public sector)
Maintain internal quality procedures for high-risk AI deploymentArt.17Deployer (acts as provider)
Document corrective actions taken in response to AI system issuesArt.20Deployer
Record all serious incidents and near-missesArt.73Deployer
Maintain updated risk assessment as AI system usage evolvesArt.9Provider / Deployer

Domain 6: Incident Response and Reporting (Art.73, Art.20, Art.74)

Art.73 is the most operationally demanding supply chain obligation. When a high-risk AI system produces a serious incident — defined as one that causes or could cause death, serious health effects, fundamental rights violations, or significant property damage — deployers must report to national competent authorities.

The correct timelines under Art.73 are:

These are days, not hours. Do not confuse with NIS2 timelines (4h/24h/1 month) or DORA (4h/24h/72h/1 month) — the EU AI Act Art.73 uses 2/10/15-day windows.

Checklist ItemArticleOwner
Define what constitutes a serious incident for your AI deploymentArt.73(1)Deployer
Establish internal escalation path: AI incident detected → responsible team → legalArt.73Deployer
Document your national competent authority (NCA) contact in your jurisdictionArt.73(2)Deployer
Prepare 2-day initial notification template (incident facts, scope, system affected)Art.73(2)Deployer
Prepare 10-day preliminary report templateArt.73(3)Deployer
Prepare 15-day full report template with RCA sectionArt.73(4)Deployer
Establish obligation to receive vendor incident notifications within 2 daysArt.73 / Art.13Provider → Contract
Test incident response flow in tabletop exercise before August 2026Art.17Deployer
Document near-miss events even when no report is legally requiredArt.20Deployer
Coordinate incident reporting with NIS2/DORA obligations if overlap existsArt.73 + NIS2/DORADeployer

The 30-Day Action Plan Before August 2, 2026

With the deadline approaching, prioritize in this sequence:

Week 1 (June 2–8): Role and Risk Assessment

Week 2 (June 9–15): Vendor Audit

Week 3 (June 16–22): Documentation and Controls

Week 4 (June 23–30): Incident Response Readiness

Final Days (July 1–August 2)


What Happens If You Miss the Deadline?

The EU AI Act penalties under Art.99 are tiered:

Deployer violations — including failure to implement Art.26 safeguards, failure to report serious incidents under Art.73, or failure to maintain documentation — fall primarily under the second tier.

More practically: national competent authorities will initially focus on high-profile cases and repeat violations. But the regulatory machinery is in motion, and the documentation burden means that companies without records are disproportionately exposed when an incident triggers an investigation.


EU-Native Infrastructure as a Compliance Accelerant

One underappreciated supply chain risk: US-headquartered AI providers operating under the CLOUD Act create a structural audit risk. If your AI vendor's logs, training data, or model weights are subject to US law enforcement access, that access may occur without your knowledge — and may constitute a data breach under GDPR Art.4(12) that itself triggers NIS2 or DORA reporting obligations, in addition to Art.73.

EU-native infrastructure eliminates this category of risk. Running AI workloads on EU-hosted platforms (Hetzner, Scaleway, OVHcloud, or managed platforms like sota.io) ensures that your AI system's operating environment is subject to EU law exclusively — simplifying Art.26 documentation and reducing the surface area for cross-jurisdictional incident scenarios.


Series Summary: EU AI Act AI Supply Chain 2026

This five-part series has covered the full scope of supply chain compliance:

  1. When SaaS Developers Become Deployers — the Art.3 classification decision, provider vs. deployer, and the 35-item initial checklist
  2. Supply Chain Due Diligence — Art.13 & Art.26 — what documentation you must obtain from vendors and what you must verify before deploying
  3. AI Supply Chain Contracts — Vendor Requirements — the specific provisions that must appear in vendor agreements to satisfy Art.25
  4. Incident Response — Art.73 Deployer Obligations — the 2/10/15-day reporting framework and the 25-item incident response checklist
  5. This post — the complete 60-item master checklist across all six compliance domains

The August 2, 2026 deadline is firm. The checklist above is actionable. Start with role classification this week, and work through each domain in sequence. The SaaS teams that are prepared will not just avoid penalties — they'll be positioned to answer enterprise procurement questionnaires and EU public sector tenders that now require AI Act compliance attestations.

EU-Native Hosting

Ready to move to EU-sovereign infrastructure?

sota.io is a German-hosted PaaS — no CLOUD Act exposure, no US jurisdiction, full GDPR compliance by design. Deploy your first app in minutes.