By Heath Muchena · Sources reviewed 2026-11-13 · Original DN analysis and explicitly illustrative examples
What Matters
A smart-contract feature is not a general AI Act compliance certificate. Identify the exact rule, actor, activity and document status before evaluating the implementation.
- AI Act and Data Act claims must not be merged by name.
- A proposed amendment is not an enacted outcome.
- Applicability reasoning and control evidence are separate records.
The DN thesis
The useful deliverable is a claim that a reviewer can inspect and challenge. DN’s map preserves the legal-source question, the scope question and the implementation question without turning document completeness into a compliance score.
1. Which law and exact provision does the claim mean?
“AI Act compliant smart contract” is too compressed to review. It can mix a software component, an AI system, an organisation’s role and several unrelated duties into one label. Start by replacing that label with a sentence: this actor, carrying out this activity, is said to meet this particular requirement under this named legal text.
A useful correction is the distinction between the AI Act and the Data Act. The Data Act’s Article 36 concerns smart contracts used to execute data-sharing agreements and identifies specified vendors or professional deployers. Its controls include robustness, access controls, termination or interruption, archiving and consistency with the agreement. [1] That is not a universal list of clauses that the AI Act requires in every contract.
Record the provision and its actual version. A source title containing “AI,” an audit badge or a supplier’s compliance landing page is not a substitute for identifying the rule being claimed. One project may need more than one legal analysis; use a separate record for each claim rather than pretending one checkbox covers all of them.
The instrument below performs one deliberately narrow check: whether the two named requirement families are attributed to the corresponding instrument. It does not read the provision or decide that the selected law applies.
2. Whose role and intended use are being assessed?
A contract address does not identify every responsible actor. Map the organisation placing a system on the market, the organisation using it, the party deploying a data-sharing application and the people operating the relevant controls. Several roles may coexist, and the correct description can change when an organisation changes a system or its intended use.
The Commission’s AI Act overview distinguishes providers and deployers, among other actors. [2] Its classification guidance begins with whether the software is an AI system and its intended purpose, then considers the relevant high-risk routes and exceptions. [3] Simply putting an AI-generated instruction into a smart contract does not complete that analysis.
Write down the concrete use. “Agent infrastructure” tells a reviewer much less than a scoped description of what decisions are made, whose interests are affected, which component executes them and which organisation controls it. Keep that description stable while assessing the claim; otherwise an answer about one use can quietly become a badge for a different use.
DN’s agent-contract authority analysis addresses who can bind whom. This record asks a different question: whether the stated compliance claim has an identifiable actor, activity and legal basis. Neither a wallet signature nor an internal role label settles the legal role by itself.
3. Is the source enacted law, a proposal or guidance?
Currentness is a separate evidence problem. A technically accurate quotation can still be taken from a superseded version, a draft proposal or explanatory guidance. Capture the source’s status as well as its text, and record the date on which that status was checked.
This is material for smart contracts. The Commission’s Digital Omnibus working document proposes deleting Data Act Article 36. [4] At this article’s review on 13 November 2026, the European Parliament’s official procedure file for 2025/0360(COD) still displayed “Awaiting committee decision.” [5] A proposal to remove a duty is not evidence that the duty has already been removed.
The record therefore has different states for enacted text, proposal and guidance. Even “enacted text reported” is only the reader’s source classification. It does not verify commencement, transitional provisions, territorial reach, later amendments or how a court or regulator would apply the text.
A source change should reopen the affected claim, not silently overwrite the earlier record. Keep the earlier version for history, add the new review and identify what changed. The export’s timestamp records when the worksheet was built; it cannot certify the law’s status at that moment.
4. Does applicability reasoning exist independently of the control?
A team may have built a stop function before deciding which requirement it is meant to satisfy. Another team may conclude that a particular rule is outside its scope while still needing the same control for operational reasons. Both situations are possible. A control and a legal applicability conclusion answer different questions.
Keep two references: one for why the claimed obligation applies or does not apply to this actor and activity, and another for what was implemented. “Not applicable” needs a reason just as “applicable” does. A conclusion about one provision should not become a statement that no other duties exist.
For the Data Act example, the relevant published provision also addresses conformity assessment and a declaration of conformity. [1] The presence of a pause button alone cannot stand in for that wider review. The instrument does not generate a declaration, assess conformity or decide that a contractual clause meets the text.
When applicability is unknown but an implementation reference is present, both facts remain visible. The missing reasoning does not erase completed engineering work, and completed engineering work does not answer the missing legal question. This separation helps a reviewer ask for the specific missing evidence instead of commissioning an undirected new audit.
5. What evidence would make the claim reviewable after a change?
A reviewable claim needs a path from the named requirement to the actual system and its relevant version. That path might involve a design document, an access-control test, an operational procedure, a responsibility assignment or a specialist assessment. The right evidence depends on the claim; the worksheet does not prescribe a universal package.
Record the reviewer who owns the gap and the event that should reopen the assessment. A new intended purpose, a changed actor, a contract upgrade, a different control owner or revised legal text can alter the question. Avoid copying a conclusion from one deployment into another simply because the product name is unchanged.
Missing evidence is not automatically proof of noncompliance. Equally, a complete set of aliases is not proof that the underlying documents are genuine or sufficient. The useful output is a precise record of what is asserted, what source is cited, what implementation is reported and what remains unresolved.
Use the exported record as a handoff to a qualified reviewer, with the actual documents kept in an appropriate controlled system. DN’s data-transfer evidence record and oversight-cost analysis cover adjacent questions. None is a legal safe harbor, and the local instrument below does not approve deployment.
DN Smart-Contract Rule Scope Map
Map one claimed obligation to its cited instrument and independent evidence. All source-status and applicability selections are reader assertions. Use a separate record for another claim.
Use non-sensitive aliases, never credentials, personal data or confidential documents. Blank and unknown entries stay unresolved. The tool does not open evidence references.
No current record. Load a labelled example or enter your own information.
This instrument makes no network requests and uses no persistent browser storage. Its download contains the current inputs, calculation or review findings, source references and provenance. The website’s other services remain subject to its privacy policy.
Illustrative cases
These cases are invented to expose the method’s boundaries. They are not observed customers, deployments, trading outcomes or legal findings.
Illustrative documented claim, not compliance
The selected requirement and instrument names align and references are present. The output remains a documented claim only; no legal or technical evidence is authenticated.
Illustrative wrong instrument attribution
The Data Act Article 36 claim is attributed to the AI Act. The mismatch is preserved even though the remaining fields are filled in.
Illustrative proposed change treated separately
The record labels the source as a proposal. It cannot turn a proposed amendment or repeal into an enacted outcome.
Illustrative implementation without applicability
A control-test reference remains recorded while the legal applicability question is unresolved. Technical work does not fill the missing legal reasoning.
Illustrative nonapplication without reasoning
Selecting not applicable does not close the gap. The scope and reason still need a reference.
Illustrative known conflicts with missing context
The instrument mismatch and superseded-source declaration remain visible beside the missing context and implementation evidence.
Methodology, assumptions and limits
The model compares only the named claim family and instrument labels. AI-system obligations attributed to the Data Act, or Data Act Article 36 controls attributed to the AI Act, are flagged as attribution conflicts. Other instruments remain outside that narrow comparison. This is not interpretation of a pasted clause.
Applicability, implementation evidence and source-version review are separate outputs. A superseded-source declaration or instrument mismatch remains visible when other inputs are missing. Proposals and guidance trigger review rather than becoming an enacted-duty assertion. A reported nonapplication conclusion still needs its reasoning reference.
References are exact reader-entered aliases and are not opened or authenticated. Raw inputs and example provenance are exported. Changes invalidate the displayed record and download until a new record is built. The review date describes these cited sources; an export timestamp does not refresh them.
Frequently asked questions
Does every smart contract need an AI Act kill switch?
This article does not support that blanket claim. The relevant published smart-contract provision discussed here is Data Act Article 36, with its own activity and actor scope. Applicability and later amendments require a separate review.
Does a proposal to delete Article 36 mean it is already gone?
No. A proposal and an enacted amendment are different states. The official procedure reviewed on 13 November 2026 remained pending; this record does not monitor later changes.
Can a complete record certify compliance?
No. It makes the claim and its references reviewable. It does not authenticate documents, interpret law, inspect code or assess conformity.
What if the team has built a control but applicability is unknown?
Keep the implementation reference and the unresolved applicability question separately. One does not prove the other.
Primary sources
Reviewed 2026-11-13. A later export timestamp does not update the sources.
- Data Act, Regulation (EU) 2023/2854, Article 36 — Published smart-contract requirements for the specified data-sharing context; not a universal AI Act clause list.
- European Commission: Navigating the AI Act — Roles and risk-based framework; this article does not reproduce a universal application deadline.
- AI Act Service Desk: General principles for classification — AI-system definition, intended purpose and classification routes; explanatory guidance is not an individual legal determination.
- Commission working document on the Digital Omnibus proposal — Explains the proposed deletion of Data Act Article 36; not evidence of an enacted repeal.
- European Parliament: Procedure 2025/0360(COD) — Official procedure status reviewed 13 November 2026; later developments are not monitored by the tool.
Turn an evidence gap into a focused question
For a narrowly documented source, version or scope question, explore a DN Research Desk enquiry. Research scope and availability are agreed separately; legal advice and conformity assessment are not supplied by this worksheet.
Disclosure: this article links to DN’s own research service. A commercial enquiry does not purchase a favorable conclusion. This instrument is educational documentation support and does not provide personalized financial, legal or security approval.
