Classification and role determination protocol (PR-CLS)

PR-CLS determines the role an organisation holds under Regulation (EU) 2024/1689, whether an AI system is high-risk under Art. 6 with Annex I or Annex III, whether a model is a general-purpose AI model, whether that model carries systemic risk under Art. 51, and which duties under Art. 50, Art. 53 and Art. 54 apply. It applies to any AI system or model before an obligation register exists. It produces determination records, an obligation register with application dates from data, and a classification report for discussion with counsel.

Protocol PR-CLS: identification
Field	Value
Protocol	PR-CLS
Version	v1.0
Articles	Art. 6 Art. 51 Art. 50 Art. 53 Art. 54 Art. 55
Derived from	`Commission guidelines on high-risk classification`Draft `Commission guidelines on GPAI scope`Published Annex III Article 6(3)

PR-CLS v1.0 is built from Commission guidelines on high-risk classificationDraft, Commission guidelines on GPAI scopePublished, Annex III of Regulation (EU) 2024/1689 and Art. 6(3) of Regulation (EU) 2024/1689.

Scope

The protocol covers three determinations and three applicability checks.

Role: which definitions in Art. 3 fit the organisation for a given system (provider, deployer, importer, distributor, authorised representative), whether an Art. 25(1) trigger turns a deployer or distributor into a provider, and whether Art. 2(1) brings a third-country organisation into scope.
Risk tier: whether the system is high-risk under Art. 6(1) with Annex I, or under Art. 6(2) with Annex III, and whether an Art. 6(3) condition removes the Annex III classification. The protocol checks the Art. 6(4) documentation and the Art. 49(2) registration data set that follow a not-high-risk conclusion.
General-purpose AI model status: whether a model meets Art. 3(63), whether it is presumed to have high-impact capabilities under Art. 51(2), whether the Art. 52(1) notification was made, and whether a modification makes the modifier the provider of a new model.
Applicability of Art. 50: which paragraphs apply to each function and whether each claimed exemption matches the observed behaviour.
Applicability of Art. 53(2) and Art. 54: whether the free and open-source exception holds, and whether the authorised representative mandate and custody meet Art. 54(1), (3) and (4).
The classification gate of Art. 55: the cumulative training compute calculation that decides whether Art. 55 applies.

The protocol is the entry procedure of the Classification and readiness assessment service and the scoping procedure of the GPAI evaluation support service. Every result is a technical assessment for discussion with counsel; the limits of that assessment are stated on what we cannot do.

The protocol does not cover marking and detection tests, which run under PR-50-MARK v1.0; the content of Annex XI and Annex XII documentation, which runs under PR-11-DOC v1.0; model evaluation and adversarial testing, which run under PR-15-ROB v1.0; or the high-risk requirements in Art. 9 to Art. 17, which have their own protocols.

Articles covered

Testable elements are quoted from the obligation records. Rows marked "no obligation record" state an element derived from the Act text, because Art. 6, Art. 51 and Art. 52 have no obligation page. Elements that another protocol tests are listed so that every testable element of a covered record appears once; step 41 records the hand-off.

Articles covered
Article	Testable element	Steps
`Art. 3(3)`, `(4)`, `(6)`, `(7)`, `Art. 25(1)`, `Art. 2(1)` (no obligation record)	Each `Art. 3` role definition is tested against the organisation's relation to the system, each `Art. 25(1)` trigger is checked, and the territorial scope condition in `Art. 2(1)` is checked.	8, 9, 10, 11
`Art. 6(1)`, Annex I (no obligation record)	The system is, or is a safety component of, a product listed in Annex I that needs third-party conformity assessment under the listed legislation.	12
`Art. 6(2)`, Annex III (no obligation record)	The intended purpose of each function matches, or does not match, an Annex III area and one of its listed use cases, with the consistency of the intended-purpose materials checked.	13, 14
`Art. 6(3)` (no obligation record)	For each Annex III match, one of the four `Art. 6(3)` conditions applies or none applies, and the profiling rule is checked.	15, 16
`Art. 6(4)`, `Art. 49(2)` (no obligation record)	The `Art. 6(3)` assessment is documented before placing on the market, and the registration entry carries the Annex VIII Section B data set.	17, 18
`Art. 3(63)`, `Art. 51`, `Art. 52` (no obligation record)	The model meets or does not meet the general-purpose definition, the cumulative training compute is compared with the `Art. 51(2)` presumption, and the `Art. 52(1)` notification and any `Art. 52(2)` rebuttal are recorded.	19, 20, 21, 22, 23, 24, 25, 26
Art. 50	Each function of the system is assigned to the Art. 50 paragraphs that apply to it (interaction, generation, emotion recognition or biometric categorisation, deep fake, public-interest text), and each claimed exemption under Art. 50(1), (2), (3) or (4) is checked against the observed behaviour of the function.	5, 35, 36, 37
Art. 50	A natural person who starts an interaction sees or hears that they interact with an AI system at the latest at the first interaction, in a clear and distinguishable manner, without opening a separate page or document (Art. 50(1) with Art. 50(5)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	Every synthetic output of each modality (audio, image, video, text) carries a machine-readable mark on generation, verified on a sample of outputs per modality and per output path (Art. 50(2)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	Each marked output is detectable as artificially generated or manipulated by a detection solution that a party other than the provider can run, verified by running the detection on marked samples (Art. 50(2), detectability).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	The mark survives a defined set of ordinary transformations for the modality (re-encoding, resizing, cropping, screenshot, format conversion, transcoding, copy and paste for text), verified per transformation with the survival result recorded (Art. 50(2), robustness).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	The mark is readable by tools other than the provider's own, such as a provenance manifest validator or a shared detection interface, verified against at least one independent reader (Art. 50(2), interoperability).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	The detection solution returns the correct result on marked outputs and on non-generated control samples, with the detection rate and the false-positive rate recorded against the target the provider justified (Art. 50(2), reliability and effectiveness).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	A natural person exposed to an emotion recognition or biometric categorisation system receives the notice of its operation at the latest at the first exposure, verified on the deployment interface or premises (Art. 50(3) with Art. 50(5)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	Deep fake content shown to a natural person carries a disclosure that is perceivable at the latest at the first exposure, embedded so that ordinary display does not hide it, verified on each distribution surface (Art. 50(4) with Art. 50(5)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	Published AI-generated text on matters of public interest carries a disclosure at the point of publication, unless the editorial-control exemption is recorded for that publication, verified on the published surface (Art. 50(4) with Art. 50(5)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 50	Each disclosure under Art. 50(1), (3) and (4) conforms to the applicable accessibility requirements, verified with the accessibility checks the applicable requirement prescribes (Art. 50(5)).	Tested by PR-50-MARK v1.0; hand-off at 41
Art. 53	Provider status after modification: whether a fine-tuning, continued training or other modification makes the modifier the provider of a new general-purpose AI model, determined against the criteria in the Commission guidelines on GPAI scope, including the compute used for the modification relative to the training compute of the original model	21, 22, 27; criteria from `Commission guidelines on GPAI scope`Published
Art. 53	Art. 53(2) exception status: whether the licence allows access, usage, modification and distribution, and whether the parameters including the weights, the architecture information and the usage information are publicly accessible at the stated locations	28, 29, 30
Art. 53	Annex XI completeness: every element of Section 1 point 1(a) to (f) and point 2(a) to (e), and of Section 2 for a model with systemic risk, is present in the technical documentation and traceable to a record	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 53	Annex XII completeness: every element of point 1(a) to (h) and point 2(a) to (c) is present in the information made available to downstream AI system providers	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 53	Cross-document consistency: the architecture, number of parameters, modality, format, release date, distribution method and licence match between the Annex XI documentation, the Annex XII information and the public model card	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 53	Declared input and output limits: the modality, format and maximum input and output size stated under Annex XII point 2(b) match the behaviour observed on a sample of inputs at and beyond the stated limits	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 53	Evaluation result traceability: each evaluation result cited under Art. 53(1)(a) carries the evaluation protocol name, the dataset version and the model version, and each identifier resolves to a stored record	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 53	Training-content summary structure: the published summary follows the sections of the AI Office template and each section carries content rather than a placeholder	Tested by PR-11-DOC v1.0; hand-off at 41
Art. 54	Mandate verification: the written mandate exists, appoints a representative established in the Union, carries a date earlier than the placing of the model on the Union market, lists each task in Art. 54(3)(a) to (d) and the power in Art. 54(4), and a copy in one of the official languages of the institutions of the Union is produced on request (Art. 54(1), (3) and (4))	4, 31, 32
Art. 54	Retrieval exercise: on a simulated reasoned request, the representative produces from its own custody the current Annex XI technical documentation copy, the provider's contact details and the information that shows the Chapter V obligations are fulfilled (Art. 54(3)(b) and (c))	33, 34
Art. 55	Model evaluation under Art. 55(1)(a): for each identified systemic risk, an evaluation is run with a named protocol and tool set that reflects the state of the art, and the result is recorded against a stated acceptance criterion	Tested by PR-15-ROB v1.0; hand-off at 41
Art. 55	Adversarial testing under Art. 55(1)(a): a red-team campaign with defined attack classes, elicitation methods and a success criterion per attack is executed against the model, and each outcome is recorded	Tested by PR-15-ROB v1.0; hand-off at 41
Art. 55	Mitigation effectiveness under Art. 55(1)(b): each safety mitigation is re-tested with the same evaluation and attack set, and the change from the pre-mitigation baseline is measured	Tested by PR-15-ROB v1.0; hand-off at 41
Art. 55	Model cybersecurity under Art. 55(1)(d): unauthorised access to model weights, model extraction through the serving interface, training-data poisoning and prompt injection are attempted under a documented threat model, and the result of each attempt is recorded	Tested by PR-15-ROB v1.0; hand-off at 41
Art. 55	Systemic-risk classification gate under Art. 51(2): cumulative training compute is calculated from the training records in floating point operations and compared with the presumption in Art. 51(2), with the calculation method recorded	21, 22, 23

Inputs required

Inputs required
Input	Format	Provided by
Intended-purpose statement, instructions for use, marketing materials, terms of service	PDF export or URL list, one entry per material	Client
Function inventory: each function with inputs, outputs and affected natural persons	Table (CSV or spreadsheet) on the `PR-CLS` inventory template	Client, on the protocol template
Interface access for behaviour observation	Test account or staging URL with the same functions as production	Client
Model card, licence text and distribution channel list for each model	Text files and URL list	Client
Training records: accelerator type and count, training time, utilisation, or architecture and training token counts	Spreadsheet or log export per training run	Client
Modification training records for a fine-tuned or continued-trained model	Spreadsheet or log export per modification run	Client
`Art. 6(3)` assessment document, where a not-high-risk conclusion is claimed	PDF	Client
EU database registration entry or its draft	Export or screenshot of the Annex VIII Section B fields	Client
`Art. 52(1)` notification correspondence and any `Art. 52(2)` rebuttal	Copy of the correspondence	Client
Written mandate and mandate copy in an official language of the institutions of the Union	PDF	Client or the authorised representative
Custody records of the authorised representative: Annex XI copy, provider contact details	Repository export or PDF	Authorised representative
Regulation entry ids for each application date in the register	List of ids from the regulation data; the governing entry for this version is 2026-07-27In force	Protocol, generated

Steps

Preparation

Record the system name, version, provider name and place of establishment.
Record System identification record: name, version, provider, place of establishment
[AIAssuranceLabs procedure]
Collect the intended-purpose statement, instructions for use, marketing materials and terms of service for the system.
Record Intended-purpose evidence set with one entry per material
[Commission guidelines on high-risk classification, draft, general principles, para. 12 on intended purpose]
Collect the licence text, model card, distribution channel list and training records for each model.
Record Model evidence set with one entry per model
[AIAssuranceLabs procedure]
Collect the written mandate and the custody records of the authorised representative where the provider is outside the Union.
Record Mandate evidence set
[Regulation (EU) 2024/1689, Art. 54(1) and (3)]
List each function of the system with its inputs, outputs and the natural persons it affects.
Record Function inventory with a function id per row
[AIAssuranceLabs procedure]
Record the interface access used to observe each function during the exemption checks.
Record Access record: interface, account, functions reachable
[AIAssuranceLabs procedure]
Record the regulation entry id for each application date the obligation register will carry.
Record Entry reference list
[AIAssuranceLabs procedure]

Execution

Check whether the organisation develops the system, or has it developed, under its own name or trademark.
Record Role determination record, provider test
[Regulation (EU) 2024/1689, Art. 3(3)]
Check whether the organisation uses, imports or distributes the system, and record each Art. 3 role that fits.
Record Role determination record, deployer, importer and distributor tests
[Regulation (EU) 2024/1689, Art. 3(4), (6) and (7)]
Check each Art. 25(1) trigger: own name on a high-risk system, substantial modification, change of intended purpose.
Record Art. 25 trigger record
[Regulation (EU) 2024/1689, Art. 25(1)(a) to (c)]
Check whether the provider is established outside the Union and whether the output is used in the Union.
Record Territorial scope record
[Regulation (EU) 2024/1689, Art. 2(1)(a) and (c)]
Check whether the system is an Annex I product or safety component that needs third-party conformity assessment.
Record Annex I route record: product, safety component status, conformity assessment route
[Regulation (EU) 2024/1689, Art. 6(1)(a) and (b); Commission guidelines on high-risk classification, draft, Annex I guidance]
Match the intended purpose of each function against the Annex III areas and their listed use cases.
Record Annex III matching record, one row per function
[Regulation (EU) 2024/1689, Art. 6(2) and Annex III points 1 to 8; Commission guidelines on high-risk classification, draft, Annex III guidance]
Check whether all materials limit the system's uses consistently, or leave a high-risk use case open.
Record Intended-purpose consistency record
[Commission guidelines on high-risk classification, draft, general principles, para. 12]
For each Annex III match, test the function against the four Art. 6(3) conditions and record which applies.
Record Art. 6(3) condition record, one row per Annex III match
[Regulation (EU) 2024/1689, Art. 6(3) first subparagraph (a) to (d); Commission guidelines on high-risk classification, draft, Annex III guidance, section on Art. 6(3)]
Check whether the function performs profiling of natural persons, which removes the Art. 6(3) exception.
Record Profiling check record
[Regulation (EU) 2024/1689, Art. 6(3) third subparagraph; Commission guidelines on high-risk classification, draft, Annex III guidance]
Check that the provider documented the Art. 6(3) assessment before placing on the market or putting into service.
Record Art. 6(4) documentation check
[Regulation (EU) 2024/1689, Art. 6(4); Commission guidelines on high-risk classification, draft, paras. 113 to 116]
Compare the EU database entry, or its draft, with Annex VIII Section B points 1 to 6 and 8.
Record Registration data-set check, one row per Annex VIII Section B point
[Regulation (EU) 2024/1689, Art. 49(2) and Annex VIII Section B]
Check whether each model shows significant generality and performs a wide range of distinct tasks.
Record GPAI determination record, generality test
[Regulation (EU) 2024/1689, Art. 3(63); Commission guidelines on GPAI scope, published, section 2.1]
Compare the training compute with the section 2.1 indicative criterion and record the generality condition met.
Record GPAI determination record, indicative criterion and generality condition
[Commission guidelines on GPAI scope, published, section 2.1]
Calculate the cumulative training compute in floating point operations from the training records.
Record Training compute determination record, calculation
[Regulation (EU) 2024/1689, Art. 51(2); Commission guidelines on GPAI scope, published, Annex A.2]
Record the estimation approach, hardware-based or architecture-based, and the error margin of the estimate.
Record Training compute determination record, approach and error margin
[Commission guidelines on GPAI scope, published, Annex A.2.1 and A.2.2]
Compare the compute figure with the Art. 51(2) presumption and record the classification outcome.
Record Systemic-risk classification record
[Regulation (EU) 2024/1689, Art. 51(1)(a) and (2)]
Check for a Commission designation decision under Art. 51(1)(b) against the Annex XIII criteria.
Record Designation check record
[Regulation (EU) 2024/1689, Art. 51(1)(b) and Annex XIII]
Record the date the presumption was met and the date of the Art. 52(1) notification.
Record Notification record: date the presumption was met, notification date, interval
[Regulation (EU) 2024/1689, Art. 52(1); Commission guidelines on GPAI scope, published, section 2.3]
Where the provider rebuts the presumption, record each argument against the section 2.3.3 criteria.
Record Rebuttal record, one row per argument
[Regulation (EU) 2024/1689, Art. 52(2); Commission guidelines on GPAI scope, published, section 2.3.3]
For a modified model, compare the modification compute with one third of the original, or the section 3.2 substitute base.
Record Provider determination record for a modified model: both compute figures, fraction, result
[Commission guidelines on GPAI scope, published, section 3.2]
Read the licence and check that it allows access, usage, modification and distribution without monetisation conditions.
Record Open-source exception assessment record, licence test
[Regulation (EU) 2024/1689, Art. 53(2); Commission guidelines on GPAI scope, published, section 4.2]
Retrieve the weights, architecture information and usage information from the stated public locations.
Record Open-source exception assessment record, public availability test with retrieval results
[Regulation (EU) 2024/1689, Art. 53(2) and Art. 54(6); Commission guidelines on GPAI scope, published, section 4]
Record whether a systemic-risk classification removes the Art. 53(2) and Art. 54(6) exceptions.
Record Open-source exception assessment record, systemic-risk check
[Regulation (EU) 2024/1689, Art. 53(2) last sentence and Art. 54(6)]
Check the mandate: Union establishment, date before placing on the market, tasks (a) to (d), Art. 54(4) power.
Record Mandate verification record, one row per Art. 54 element
[Regulation (EU) 2024/1689, Art. 54(1), (3) and (4)]
Request the mandate copy in an official language of the institutions of the Union and record the response.
Record Mandate copy request record: language, response
[Regulation (EU) 2024/1689, Art. 54(3)]
Send a simulated reasoned request and record what the representative produces from its own custody.
Record Retrieval exercise record: items requested, items produced, source of each
[Regulation (EU) 2024/1689, Art. 54(3)(b) and (c)]
Compare the version of the retrieved Annex XI copy with the provider's current technical documentation.
Record Retrieval exercise record, version match
[Regulation (EU) 2024/1689, Art. 54(3)(b)]
Assign each function to the Art. 50 paragraphs that apply to it and record the assignment.
Record Article 50 applicability record, one row per function
[Regulation (EU) 2024/1689, Art. 50(1) to (4); Commission guidelines on Article 50 transparency obligations, published]
Observe each function with a claimed Art. 50 exemption and compare the behaviour with the exemption condition.
Record Article 50 exemption observation record, one row per claimed exemption
[Regulation (EU) 2024/1689, Art. 50(1), (2), (3) and (4); Commission guidelines on Article 50 transparency obligations, published]
Record whether a generating system was on the market before the Art. 50 application date.
Record Art. 111(4) legacy status record
[Regulation (EU) 2024/1689, Art. 111(4)]

Analysis

Where routes give different results, record the strictest applicable classification and the reason.
Record Classification resolution record
[AIAssuranceLabs procedure]
Compile the obligation register: each applicable article per role, with the regulation entry id for its date.
Record Obligation register
[AIAssuranceLabs procedure]
Record the provider's justification for each judgement the Act and the guidelines leave open.
Record Determination rationale record, one entry per open judgement
[AIAssuranceLabs procedure]
List each testable element that another protocol tests and name that protocol in the register.
Record Protocol hand-off list
[AIAssuranceLabs procedure]

Reporting

Write the classification and role determination report in the fixed record layout.
Record Classification and role determination report
[AIAssuranceLabs procedure]
Mark each determination as a technical assessment for discussion with counsel.
Record Counsel note on each determination
[AIAssuranceLabs procedure]
Deliver the report, the register and the open-items list as one versioned evidence pack.
Record Evidence pack index
[AIAssuranceLabs procedure]

The records use fixed layouts. Three are shown with synthetic values.

High-risk classification record, synthetic values

record: high-risk-classification
system: synthetic-recruitment-screening-assistant
function: F-02, ranks applications for interview
route_art_6_1: not applicable, no Annex I product
route_art_6_2: Annex III point 4(a), recruitment or selection of natural persons
intended_purpose_consistency: high-risk use left open in marketing materials
art_6_3_condition_tested: (a) (b) (c) (d)
art_6_3_condition_met: none
profiling_of_natural_persons: yes, ranks on inferred traits
result: high-risk under Art. 6(2)
resolution_ref: CRR-01
rationale_ref: DRR-03

Training compute determination record, synthetic values; arithmetic follows Annex A.2.1 of the published GPAI scope guidelines

record: training-compute-determination
model: synthetic-model-A
approach: hardware-based (Annex A.2.1)
accelerator_count: 4096
peak_flop_per_second_per_accelerator: 9.9e14
training_time_hours: 1200
utilisation_rate: 0.40
cumulative_training_compute_flop: 7.0e24
error_margin_recorded: yes
art_51_2_presumption: not met
generality_condition: generates language
section_2_1_indicative_criterion: met
art_52_1_notification: not applicable
result: general-purpose AI model, no systemic-risk presumption

Obligation register, two synthetic rows

article        | role                      | applies_to        | entry_id                  | tested_by
Art. 50(1)     | provider                  | function F-01     | article-50-transparency   | PR-50-MARK
Art. 53(1)(a)  | provider, modified model  | model M-01        | gpai-obligations          | PR-11-DOC

Thresholds and criteria

Regulation (EU) 2024/1689 sets one numeric figure in the scope of this protocol. Art. 51(2) presumes high-impact capabilities when the cumulative amount of computation used for training, measured in floating point operations, is greater than 10^25; Art. 51(3) lets the Commission amend that figure by delegated act. Step 23 records the comparison. Every other criterion in the protocol is qualitative or comes from a Commission guideline.

No harmonised standard is cited in the Official Journal for classification, so level 1 of the threshold hierarchy is empty; the standards page tracks the count. No JTC 21 draft covers classification or role determination, so levels 2 and 3 are empty. The criteria in use sit at level 4, Commission guidelines, and at level 6, the provider's own justification.

Thresholds and criteria
Determination	Criterion	Source and stage	Step
Role	The organisation matches an `Art. 3` definition; an `Art. 25(1)` trigger is present or absent	Regulation (EU) 2024/1689, `Art. 3(3)`, `(4)`, `(6)`, `(7)`, `Art. 25(1)`	8 to 11
Annex I route	Annex I product or safety component, and third-party conformity assessment under the listed legislation	Regulation (EU) 2024/1689, `Art. 6(1)`; `Commission guidelines on high-risk classification`Draft, Annex I guidance	12
Annex III route	Intended purpose matches an Annex III use case; materials that leave a high-risk use open count toward the match	Regulation (EU) 2024/1689, `Art. 6(2)`, Annex III; `Commission guidelines on high-risk classification`Draft, general principles, para. 12	13, 14
`Art. 6(3)` exception	One of the four conditions applies, and the function performs no profiling of natural persons; the conditions are read narrowly	Regulation (EU) 2024/1689, `Art. 6(3)`; `Commission guidelines on high-risk classification`Draft, Annex III guidance	15, 16
General-purpose AI model	Training compute greater than `10^23` floating point operations and the model can generate language, text-to-image or text-to-video; the criterion is indicative and rebuttable in both directions	`Commission guidelines on GPAI scope`Published, section 2.1	19, 20
Systemic-risk presumption	Cumulative training compute greater than `10^25` floating point operations	Regulation (EU) 2024/1689, `Art. 51(2)`, amendable under `Art. 51(3)`	21 to 23
Notification	The provider notifies the Commission within two weeks after the presumption is met or after it becomes known that it will be met	Regulation (EU) 2024/1689, `Art. 52(1)`; `Commission guidelines on GPAI scope`Published, section 2.3	25
Modifier as provider	Modification compute greater than one third of the original model's training compute; where the original is unknown, one third of the section 2.1 figure, or of the `Art. 51(2)` figure for a model with systemic risk; indicative	`Commission guidelines on GPAI scope`Published, section 3.2	27
Free and open-source exception	Licence allows access, usage, modification and distribution; no monetisation condition; weights, architecture information and usage information retrievable at the stated locations; no systemic-risk classification	Regulation (EU) 2024/1689, `Art. 53(2)`, `Art. 54(6)`; `Commission guidelines on GPAI scope`Published, section 4	28 to 30
Mandate	Each element in `Art. 54(1)`, `(3)(a)` to `(d)` and `(4)` is present; the copy is produced in an official language of the institutions of the Union	Regulation (EU) 2024/1689, `Art. 54`	31, 32
Retrieval exercise	Each requested item is produced from the representative's own custody at the provider's current version; no published time target exists	Regulation (EU) 2024/1689, `Art. 54(3)(b)` and `(c)`; provider justification at level 6	33, 34, 40
`Art. 50` exemptions	The observed behaviour matches the exemption condition: obvious interaction, assistive editing or no substantial alteration, ancillary biometric categorisation outside the Art. 3(40) definition, artistic work, editorial control	Regulation (EU) 2024/1689, `Art. 50(1)` to `(4)`; `Commission guidelines on Article 50 transparency obligations`Published	36

Where the Act and the guidelines leave a judgement to the provider, such as the reading of "obvious" under Art. 50(1), the acceptable retrieval time for the Art. 54 exercise, or a rebuttal of the Art. 51(2) presumption, step 40 records the provider's justification and its source. The protocol attributes no figure to the Act other than the Art. 51(2) figure. The Art. 6 criteria come from a guideline that is at draft stage under consultation; the protocol names the stage on every citation.

Outputs

Artifacts are named per the artifact naming rule. The deliverable column names the service deliverable each artifact feeds; the deliverables page specifies each deliverable with a synthetic sample.

Outputs
Artifact	Contents	Deliverable
Role determination record (`PR-CLS` output)	`Art. 3` definitions tested, `Art. 25(1)` triggers, territorial scope, result per system	Classification memo: role, in Classification and readiness assessment
High-risk classification record (`PR-CLS` output)	Annex I route, Annex III matching per function, intended-purpose consistency, result	Classification memo: risk tier, in Classification and readiness assessment
Article 6(3) assessment record (`PR-CLS` output)	Condition tested per Annex III match, profiling check, `Art. 6(4)` documentation check, registration data-set check	Classification memo: Article 6(3) rationale, in Classification and readiness assessment
Training compute determination record (`PR-CLS` output)	Compute figure, estimation approach, error margin, `Art. 51(2)` comparison, designation check, notification and rebuttal records	Annex XI technical documentation contribution, in GPAI evaluation support
Provider determination record for a modified or fine-tuned model (`PR-CLS` output)	Original and modification compute, fraction, substitute base where used, resulting role	Annex XI technical documentation contribution, in GPAI evaluation support; Classification memo: role, in Classification and readiness assessment
Open-source exception assessment record (`PR-CLS` output)	Licence test, monetisation check, retrieval results for weights, architecture and usage information, systemic-risk check	Classification memo: role, in Classification and readiness assessment
Article 54 applicability and exception determination memo (`PR-CLS` output)	Place of establishment, placing on the market, mandate verification, mandate copy request, retrieval exercise, version match	Classification memo: role, in Classification and readiness assessment
Article 50 applicability and exemption determination record (`PR-CLS` output)	Function-to-paragraph assignment, exemption observations, `Art. 111(4)` legacy status	Obligation register with applicability dates, in Classification and readiness assessment
Obligation register (`PR-CLS` output)	Applicable articles per role, regulation entry id per date, protocol hand-off per testable element, open items	Obligation register with applicability dates, and Gap list against the applicable articles, in Classification and readiness assessment
Determination rationale record (`PR-CLS` output)	Provider justification and source for each judgement without a published threshold	Classification memo, in Classification and readiness assessment
Classification and role determination report (`PR-CLS` output)	All records above, the classification resolution record, and the counsel note on each determination	Classification memo, in Classification and readiness assessment; scoping input to GPAI evaluation support

Limitations

The protocol records a technical assessment of classification and role against the Act text and the Commission guidelines. It does not give a legal opinion on classification, and a market surveillance authority, the AI Office or a court can reach a different result.

The exemption observations in step 36 cover the functions and the interface named in the access record. Behaviour outside that interface, or a function added after the observation, is not observed.

The training compute calculation in steps 21 to 23 depends on the training records the provider supplies. The protocol checks the estimation approach and the arithmetic; it cannot re-run training or verify hardware utilisation independently.

A Commission designation under Art. 51(1)(b), a delegated act under Art. 51(3), or a final version of the draft high-risk classification guidelines can change a recorded result. The revision history records each such change to the protocol; a determination made under an earlier version is re-run under the new one.

The registration duty that follows an Art. 6(3) conclusion has an unsettled application date. Step 18 checks the data set under the working reading; the register carries both readings.

Step 35 decides which Art. 50 paragraphs apply to a function. It does not decide whether a marking technique is adequate under Art. 50(2); that point is unsettled, and the marking tests run under PR-50-MARK v1.0.

The protocol does not test the content of Annex XI or Annex XII documentation, the presence or survival of a mark, or the evaluation and adversarial testing of a model with systemic risk. Step 41 records which protocol tests each of those elements.

Sources

Commission guidelines on high-risk classificationDraft: draft Commission guidelines on the classification of high-risk AI systems under Article 6, under consultation, in three parts: general principles, Annex I guidance and Annex III guidance.
Commission guidelines on GPAI scopePublished: guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act, with Annex A on estimating training compute.
Commission guidelines on Article 50 transparency obligationsPublished: guidelines on the implementation of the transparency obligations for certain AI systems under Article 50, used for the exemption conditions in steps 35 and 36.
Regulation (EU) 2024/1689, consolidated text on EUR-Lex: https://eur-lex.europa.eu/eli/reg/2024/1689/oj, for Art. 2, Art. 3, Art. 6, Art. 25, Art. 49, Art. 50, Art. 51, Art. 52, Art. 53, Art. 54, Art. 111 and Annexes I, III, VIII and XIII.

Revision history

Each version of PR-CLS names the regulation entry whose verified-on date anchors the standards and guidelines it was written from.

Revision history of PR-CLS
Version	Note	Anchor
v1.0	Initial version, written from the standards in the derivation line as they stood on the verified-on date of the entry.	Verified on 18 September 2026K&L Gates, EU Digital Omnibus on AI Enters Into Force (31 July 2026)