Skip to content

Reference and glossary

ItemValue
RegulationEU 2024/1689
Snapshot16 July 2026
Documentation source check21 July 2026
MethodologyEUAIA-2026-07-16-defensible-system-scope
Categories11
Parent requirements34
Atomic items72
Assessment unitNamed AI system
CodeMeaning
SCOPEScope, definition, role, timing, literacy
PROHIBITEDArticle 5 hard-stop screen
CLASSIFICATIONArticle 6, Annex I and Annex III
HIGHREQArticles 8–15
OPERATORSOperator and conformity duties
FRIAArticle 27
TRANSPARENCYArticle 50
MONITORINGArticles 72–73
GPAI_MONITORINGArticle 55
GPAIChapter V model and supply-chain duties
NAGoverned non-applicability trail
DimensionValues
Legal answerCompliant, Non-compliant, N/A, Unassessed
Depth3 Assured, 2 Supported/Partial, 1 Asserted, 0 Gap
ConfidenceLow, Medium, High
Legal stateIn force, Future obligation, Proposed change only
StateMeaning
not_assessedNo route decision
applicableTriggered
not_applicable_with_evidenceExcluded through an approved decision
incompleteFacts or decision missing
implementedRouted work addressed, subject to assurance
prohibited_stopArticle 5 stop
future_obligationFuture readiness
proposed_change_onlyNot binding law
DateSignificance
1 August 2024Entry into force
2 February 2025AI literacy and prohibited practices apply
2 August 2025GPAI provider and governance provisions apply
2 August 2026General application date in the adopted Regulation, subject to exceptions/amendments
2 August 2027Original Annex I product-system high-risk date

Check current formally adopted law before relying on the table.

Provider

Entity developing or having developed and placing a system/model on the market or putting it into service under its name or trademark.

Deployer

Entity using an AI system under its authority in a professional context.

Importer

Union-established entity placing a third-country provider’s high-risk system on the Union market.

Distributor

Supply-chain entity making a system available without being provider or importer.

Product manufacturer

Manufacturer placing a product containing or using the high-risk system on the market under its name or trademark.

Authorised representative

Union-established person mandated to perform specified provider tasks. Article 22 and Article 54 routes must be distinguished.

GPAI model provider

Provider placing a general-purpose AI model on the market.

Downstream provider

Provider integrating a model into an AI system.

Applicable

The legal trigger and role are present.

Atomic item

Smallest independently answered legal check in Gamut.

Assessment basis

The material legal-routing facts to which confirmation applies. A material change invalidates prior confirmation.

Current-law conclusion

Position using applicable provisions in force as of the stated snapshot.

Future readiness

Preparation for applicable obligations not yet used as current law.

Hard stop

Condition preventing confirmation regardless of other scores; Article 5 is the principal example.

N/A

Approved conclusion that a specific legal trigger does not apply. It is not “not yet done.”

Assured

Depth 3: current implementation, accepted evidence, effective test and no unresolved adverse finding.

Confirmation

Accountable human sign-off of the current validated assessment basis. It is not certification, deployment approval or risk acceptance by itself.

FRIA

Fundamental Rights Impact Assessment under Article 27 for specified deployers and uses.

GPAI

General-purpose AI model. Model-provider duties and downstream system duties are distinct.

Profiling override

Article 6 logic preventing reliance on the Annex III exception where profiling of natural persons is performed.

Substantial modification

Change that can trigger provider-role transfer and renewed compliance work under Article 25.

Are there six EU AI Act risk classes in Gamut?

Section titled “Are there six EU AI Act risk classes in Gamut?”

No. Earlier wording used six presentation groupings. The defensible method uses eleven orthogonal routes because scope, prohibited practices, high-risk duties, transparency and GPAI can coexist.

Does “not high-risk” mean out of scope?

Section titled “Does “not high-risk” mean out of scope?”

No. Scope, Article 5, Article 4, Article 50 and other duties may still apply.

Does using a GPAI model make our system a GPAI model?

Section titled “Does using a GPAI model make our system a GPAI model?”

Not necessarily. Determine whether the organisation is a GPAI model provider, downstream system provider, deployer or API consumer.

Does human review mean an Annex III system is not high-risk?

Section titled “Does human review mean an Annex III system is not high-risk?”

No. Analyse the intended use, material influence, Article 6 exception and profiling. A nominal human-in-the-loop is not a blanket exclusion.

Does every high-risk deployer need a FRIA?

Section titled “Does every high-risk deployer need a FRIA?”

No. Article 27 has its own deployer and use triggers.

Can a control be Compliant but not Assured?

Section titled “Can a control be Compliant but not Assured?”

Yes. Compliant is the legal answer; depth describes verification. The claim may be Supported or Asserted until evidence and testing are complete.

No. Use future-obligation status. N/A means the legal trigger does not apply.

Can an average offset one non-compliant obligation?

Section titled “Can an average offset one non-compliant obligation?”

No. Category percentages support progress tracking; they do not erase an atomic legal gap.

No. It provides system-scoped advisory analysis. Human assessors accept evidence, approve N/A, resolve legal issues and confirm.

What happens when the selected system changes?

Section titled “What happens when the selected system changes?”

The assessment and AI context switch to that system. Prior AI analysis is cleared to prevent cross-system reuse.

No. Identity, organisation, workspace, role, plan and object-scope access controls remain required.

No. It means the assessment passed Gamut’s confirmation gates under its stated scope and snapshot.

Gamut assesses the EU AI Act per named AI system. It first establishes EU scope, intended purpose and every operator role. Gamut then calculates eleven legal routes, including an eight-part Article 5 hard-stop screen, Annex I and all eight Annex III points, operator duties, FRIA, Article 50 and GPAI. Broad provisions are decomposed into 72 atomic checks. Each applicable item receives a legal answer and a separate assurance depth. Confirmation requires accepted evidence, passing testing and no unresolved adverse finding for every applicable in-force item, while future and proposed changes are reported separately. AI can assist analysis but cannot change routing or approve the conclusion.