# EU AI Act Article 6: classify contract-AI tools by purpose, not vendor claims

Natalie Fletcher · November 4, 2026

> EU AI Act Article 6 classifies contract-review AI by purpose, not vendor claims, to assign Annex III high-risk status for tools affecting essential entitlements.

This guide shows how EU AI Act Article 6 classification, not vendor marketing claims, determines the Annex III high-risk conformity assessment path.

It focuses on contract-review tools that score eligibility for public-service or essential-private entitlements.

![This guide shows how EU AI Act Article — EU AI Act Article 6](https://static.mm-ais.com/article-images-ai/eu-ai-act-article-6-classify-contract-ai-ai-4d3e2deb.jpg)

## How Article 6 classifies contract AI

Under the EU AI Act, Article 6(1) is the gate: a system becomes subject to the Chapter III, Section 2 requirements only when it is classified as high-risk by that article. As defined by Article 3(19) and explained by artificialintelligenceact.eu/Article/29, "conformity assessment" is the process of demonstrating whether those Chapter III, Section 2 requirements are met. For contract AI, the classification turns on intended purpose, not model architecture. A transformer used solely to draft boilerplate clauses and a rules engine used to score loan-application contracts can land in different Article 6 buckets, so vendor marketing claims cannot substitute for that purpose-based check.

For contract-analysis tools specifically, the trigger to isolate is the Annex III point 4/5(a)-style eligibility triggers that pull contract-analysis systems into high-risk scope. If the tool evaluates a natural or legal person's eligibility for a public-service or essential-private entitlement, such as credit, insurance, or admission decisions, the intended purpose points toward Annex III coverage rather than a purely informational use. Your check must be strict: does the output determine or materially score eligibility, or does it only summarize terms? The former is the purpose that can cross into high-risk classification under Article 6(2), and it is that purpose statement in your technical documentation that must be tested against Annex III, not the sales page.

Timing is a separate check, not a reason to delay classification. Per artificialintelligenceact.eu/Article/113/#p-article-113-0.b, Article 113(b) is cited as the provision stating when the conformity assessment obligations come into force (2 August 2025 in that text). However, the operative fact for classification is the intended purpose at the time of placing on the market or putting into service. Do not wait for that point to run the Article 6(2)/Annex III classifier on every contract-AI tool in your stack, as the decision to classify must be made against intended purpose regardless of when the demonstration obligations apply.

Finally, if classification places the tool in high-risk scope, the obligation is to be able to demonstrate conformity through the process described in Article 3(19) and Article 29. The classification itself is the decision point that forces the need for an Annex III high-risk conformity assessment. Treat that decision point as non-delegable to the vendor, and treat your documented intended purpose as the evidence that must hold up to that test.
![How Article 6 classifies contract AI — EU AI Act Article 6](https://static.mm-ais.com/article-images-ai/eu-ai-act-article-6-classify-contract-ai-ai-27ce42d9.jpg)

## The evidence: what the sources show

Across the reviewed material, the enforcement point is not the vendor's marketing claim, but the process that authorizes who can actually perform the check. According to artificialintelligenceact.eu on Article 29, conformity assessment is formally defined as the process of demonstrating whether the requirements set out in Chapter III, Section 2 relating to a high-risk AI system are met. With that definition in mind, the first practical check for any contract-review tool is purpose-first: take its stated purpose and check it against Annex III before accepting any self-classification. If the purpose does not match, the Chapter III, Section 2 demonstration is not triggered by that list, but if it does match, the demand for that demonstration becomes real.

This section alone presents the multi-source convergence on designation supervision as the enforcement lever. From howtoregulate.org's deep read on harmonised decision-making, the designation, supervision and de-designation processes are the key leverage for influencing the conformity assessment bodies' work. The sources point to the same structural fact: a conformity assessment body does not operate in a vacuum, it operates under a designating authority that can keep, supervise, or remove its designation. That procedural control is the pressure point, not a general call for "more guidance".

That leverage translates to a two-track verification for your stack. Track one is the Annex III match against the tool's exact function, using the Article 29 definition of conformity assessment as the standard for what must be demonstrated. Track two is the notified body itself: for that exact function, confirm its designation status and that it is under active supervision for it, not just that it is a notified body in general. These are different checks, and relying on only one of them leaves the vendor's branding to fill the gap.

For contract-review tools, this means the enforceable demand is procedural. Instead of asking for a blanket "compliant" statement, ask for proof tied to a designated body for the specific high-risk purpose you have identified. Because the ability to designate, supervise, or de-designate is the lever shown across these sources, pressing on that body-first track is the way to make the purpose-first classification hold weight in practice.

![The evidence: what the sources show — EU AI Act Article 6](https://static.mm-ais.com/article-images-pixabay/eu-ai-act-article-6-classify-contract-ai-bcb0fa87.jpg)

## Three compliance paths, one winner

When you must decide how to treat a contract-AI tool in your stack, only three realistic compliance paths exist. Option A is to trust vendor marketing, such as an "AI Act ready" badge. On its own, that badge carries zero verification value, because the duty does not come from the vendor's claim. Option B is to demand a declaration of conformity plus a notified-body certificate for Annex III uses. This is more concrete than Option A, but it leaves you fully dependent on the vendor's selected conformity assessment body and on that body remaining authorized for the exact use you rely on. Option C is dual verification: demand the declaration and notified-body certificate, but also build your own documented Article 6 classification memo for that exact tool and use case. This section alone contains the three-option decision table below to make that tradeoff explicit.

## Frequently Asked Questions

**Which article serves as the gate for subjecting a contract-AI tool to Chapter III, Section 2 requirements?**

Article 6(1) is the gate, as a system becomes subject to those requirements only when it is classified as high-risk by that article.

**Can two contract-AI tools with different technologies end up in different Article 6 classifications?**

Yes, a transformer used solely to draft boilerplate clauses and a rules engine used to score loan-application contracts can land in different Article 6 buckets.

**Can vendor marketing claims override the classification process for these tools?**

No, vendor marketing claims cannot substitute for the purpose-based check required under Article 6.

**Which types of eligibility decisions can push a contract-analysis tool toward Annex III coverage?**

Evaluating a natural or legal person's eligibility for a public-service or essential-private entitlement, such as credit, insurance, or admission decisions, points toward Annex III coverage.

**How is 'conformity assessment' defined for this high-risk path?**

As defined by Article 3(19), 'conformity assessment' is the process of demonstrating whether those Chapter III, Section 2 requirements are met.

**Which specific Annex III elements must be isolated when assessing contract-analysis tools?**

The trigger to isolate is the Annex III point 4/5(a)-style eligibility triggers that pull contract-analysis systems into high-risk scope.

## Quick answers

| What is the gate for applying Chapter III, Section 2 requirements? | Article 6(1) is the gate: a system becomes subject to the Chapter III, Section 2 requirements only when it is classified as high-risk by that article. |
| --- | --- |
| What determines Article 6 classification for contract AI? | For contract AI, the classification turns on intended purpose, not model architecture. |
| Can vendor marketing claims determine the Annex III high-risk path? | No, vendor marketing claims cannot substitute for the purpose-based check. |
| What evaluation can trigger Annex III coverage for contract-analysis tools? | If the tool evaluates a natural or legal person's eligibility for a public-service or essential-private entitlement, such as credit, insurance, or admission decisions, the intended purpose points toward Annex III coverage. |
| How is 'conformity assessment' defined in this context? | As defined by Article 3(19), 'conformity assessment' is the process of demonstrating whether those Chapter III, Section 2 requirements are met. |

Also worth reading: **AI Content Labels: European Union AI Act Article 50—Label by Aug. 2, 2026**: [AI Content Labels: European Union](https://lawr.io/blog/ai-content-labels-european-union-ai-act-article-50label-by-aug-2-2026.php) · **EU AI Act High-Risk Article 13: Why 0.94 F1 Fails the Audit**: [EU AI Act High-Risk Article](https://lawr.io/blog/eu-ai-act-high-risk-article-13-why-094-f1-fails-the-audit.php) · **Contract clause review: 92% recall with manual vs automated triage**: [Contract clause review: 92% recall](https://lawr.io/blog/contract-clause-review-92-recall-with-manual-vs-automated-triage.php)

### Related reading

- [Legal Impossibility vs

Force Majeure 7 Key Differences in Contract Defense Claims](https://lawr.io/blog/legal_impossibility_vs_force_majeure_7_key_differences_in_c.php)
- [How Local AI Contract Analysis Tools Are Reshaping Legal Practice Insights from 7 Macon GA Law Firms in 2024](https://lawr.io/blog/how_local_ai_contract_analysis_tools_are_reshaping_legal_pra.php)
- [Los Angeles Immigration Attorneys Adapt to AI Contract Review Tools in 2024](https://lawr.io/blog/los_angeles_immigration_attorneys_adapt_to_ai_contract_revie.php)
- [Contract Review Metrics: 2,865 Examples—Verify Before Approval](https://lawr.io/blog/contract-review-metrics-2865-examplesverify-before-approval.php)
- [Contract clause review: 92% recall with manual vs automated triage](https://lawr.io/blog/contract-clause-review-92-recall-with-manual-vs-automated-triage.php)
- [Contract clause extraction: 60-Page Master Service Agreement (MSA) Map vs Scroll](https://lawr.io/blog/contract-clause-extraction-60-page-master-service-agreement-msa-map-vs-scroll.php)

### Latest

- [Verifying California Family Code § 4320 spousal support claims in San Joaquin...](https://lawr.io/blog/verifying-california-family-code-4320-spousal-support-claims-in-san-joaquin-county.php)
- [Sacramento soft-tissue injury claims: when to reject the first offer](https://lawr.io/blog/sacramento-soft-tissue-injury-claims-when-to-reject-the-first-offer.php)
- [Cyber Incident Disclosure Rule: 2026 Item 1.05 4-Day vs Natural Language...](https://lawr.io/blog/cyber-incident-disclosure-rule-2026-item-105-4-day-vs-natural-language-processing.php)

Canonical: https://lawr.io/blog/eu-ai-act-article-6-classify-contract-ai-tools-by-purpose-not-vendor-claims.php
Markdown: https://lawr.io/blog/eu-ai-act-article-6-classify-contract-ai-tools-by-purpose-not-vendor-claims.php/index.md