Datasets:

gitmodelmujtaba
/

eu-ai-act-normative-triples

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
catalog_title: string
catalog_version: string
statutory_act: string
official_journal: string
eli_uri: string
celex: string
domains: list<item: struct<domain_id: string, domain_name: string, statutory_category: string, legal_basis: s (... 891 chars omitted)
  child 0, item: struct<domain_id: string, domain_name: string, statutory_category: string, legal_basis: string, doma (... 879 chars omitted)
      child 0, domain_id: string
      child 1, domain_name: string
      child 2, statutory_category: string
      child 3, legal_basis: string
      child 4, domain_summary: string
      child 5, case_studies: list<item: struct<case_id: string, title: string, system_id: string, statutory_tier: string, legal_b (... 744 chars omitted)
          child 0, item: struct<case_id: string, title: string, system_id: string, statutory_tier: string, legal_basis: strin (... 732 chars omitted)
              child 0, case_id: string
              child 1, title: string
              child 2, system_id: string
              child 3, statutory_tier: string
              child 4, legal_basis: string
              child 5, expected_conformity: string
              child 6, file_path: string
              child 7, statutory_quote: string
              child 8, regulatory_requirements: struct<mandatory_articles: list<item: string>, harmonized_frameworks: struct<nist_ai_rmf: list<item: (... 125 chars omitted)
                  child 0, mandatory_articles: list<item: string>
                      child 0,
...
ild 2, gdpr: list<item: string>
                          child 0, item: string
                  child 2, conformity_procedure: string
                  child 3, fine_exposure_tier: string
              child 9, auditor_guidance: struct<intended_purpose: string, common_pitfalls: string, remediation_guidance: string>
                  child 0, intended_purpose: string
                  child 1, common_pitfalls: string
                  child 2, remediation_guidance: string
              child 10, file_sha256: string
              child 11, provenance: struct<statutory_act: string, official_journal: string, eli_uri: string, celex: string, statutory_qu (... 163 chars omitted)
                  child 0, statutory_act: string
                  child 1, official_journal: string
                  child 2, eli_uri: string
                  child 3, celex: string
                  child 4, statutory_quote: string
                  child 5, statutory_quote_sha256: string
                  child 6, spec_file_sha256: string
                  child 7, prov_o_entity: string
                  child 8, author: string
                  child 9, verification_method: string
                  child 10, timestamp: timestamp[s]
enforcement_tier: string
sha256_hash: string
shacl_shape_ref: string
predicate: string
paragraph_number: string
source_text: string
modality: string
id: string
subject: string
object: string
article_number: int64
cross_references: list<item: string>
  child 0, item: string
to
{'id': Value('string'), 'article_number': Value('int64'), 'paragraph_number': Value('string'), 'subject': Value('string'), 'modality': Value('string'), 'predicate': Value('string'), 'object': Value('string'), 'source_text': Value('string'), 'sha256_hash': Value('string'), 'cross_references': List(Value('string')), 'enforcement_tier': Value('string'), 'shacl_shape_ref': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              catalog_title: string
              catalog_version: string
              statutory_act: string
              official_journal: string
              eli_uri: string
              celex: string
              domains: list<item: struct<domain_id: string, domain_name: string, statutory_category: string, legal_basis: s (... 891 chars omitted)
                child 0, item: struct<domain_id: string, domain_name: string, statutory_category: string, legal_basis: string, doma (... 879 chars omitted)
                    child 0, domain_id: string
                    child 1, domain_name: string
                    child 2, statutory_category: string
                    child 3, legal_basis: string
                    child 4, domain_summary: string
                    child 5, case_studies: list<item: struct<case_id: string, title: string, system_id: string, statutory_tier: string, legal_b (... 744 chars omitted)
                        child 0, item: struct<case_id: string, title: string, system_id: string, statutory_tier: string, legal_basis: strin (... 732 chars omitted)
                            child 0, case_id: string
                            child 1, title: string
                            child 2, system_id: string
                            child 3, statutory_tier: string
                            child 4, legal_basis: string
                            child 5, expected_conformity: string
                            child 6, file_path: string
                            child 7, statutory_quote: string
                            child 8, regulatory_requirements: struct<mandatory_articles: list<item: string>, harmonized_frameworks: struct<nist_ai_rmf: list<item: (... 125 chars omitted)
                                child 0, mandatory_articles: list<item: string>
                                    child 0,
              ...
              ild 2, gdpr: list<item: string>
                                        child 0, item: string
                                child 2, conformity_procedure: string
                                child 3, fine_exposure_tier: string
                            child 9, auditor_guidance: struct<intended_purpose: string, common_pitfalls: string, remediation_guidance: string>
                                child 0, intended_purpose: string
                                child 1, common_pitfalls: string
                                child 2, remediation_guidance: string
                            child 10, file_sha256: string
                            child 11, provenance: struct<statutory_act: string, official_journal: string, eli_uri: string, celex: string, statutory_qu (... 163 chars omitted)
                                child 0, statutory_act: string
                                child 1, official_journal: string
                                child 2, eli_uri: string
                                child 3, celex: string
                                child 4, statutory_quote: string
                                child 5, statutory_quote_sha256: string
                                child 6, spec_file_sha256: string
                                child 7, prov_o_entity: string
                                child 8, author: string
                                child 9, verification_method: string
                                child 10, timestamp: timestamp[s]
              enforcement_tier: string
              sha256_hash: string
              shacl_shape_ref: string
              predicate: string
              paragraph_number: string
              source_text: string
              modality: string
              id: string
              subject: string
              object: string
              article_number: int64
              cross_references: list<item: string>
                child 0, item: string
              to
              {'id': Value('string'), 'article_number': Value('int64'), 'paragraph_number': Value('string'), 'subject': Value('string'), 'modality': Value('string'), 'predicate': Value('string'), 'object': Value('string'), 'source_text': Value('string'), 'sha256_hash': Value('string'), 'cross_references': List(Value('string')), 'enforcement_tier': Value('string'), 'shacl_shape_ref': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

🏛️ EU AI Act Normative Deontic Triples & Knowledge Graph

Formal Symbolic Regulatory Knowledge Base & Multi-Framework Crosswalk

Regulation (EU) 2024/1689 (Artificial Intelligence Act)

📌 Executive Summary

The EU AI Act Normative Deontic Triples dataset provides a rigorous, machine-verifiable, symbolic representation of Regulation (EU) 2024/1689. Built using the knowledge engineering methodology established in gitmodelmujtaba/gdpr-normative-triples, this benchmark translates dense legal prose into formal Subject-Modality-Predicate-Object tuples grounded in Deontic Logic (OBLIGATION, PROHIBITION, PERMISSION, EXEMPTION).

Each triple is bound to an exact EUR-Lex statutory quote, anchored with a cryptographically verifiable SHA-256 hash, cross-mapped bidirectionally to NIST AI RMF 1.0, ISO/IEC 42001:2023, and GDPR (EU 2016/679), and coupled with an executable Cytoscape/NetworkX knowledge graph topology.

🏛️ 5-Tier Dataset Architecture

data/benchmarks/
├── eu_ai_act_normative_triples.json     # Tier 1: Canonical Deontic Triples with SHA-256 digests
├── eu_ai_act_knowledge_graph.json       # Tier 2: Cytoscape & NetworkX graph topology (elements.nodes/edges)
├── domain_data_dictionary.json          # Tier 3: Controlled legal taxonomy, actor roles & risk tiers
├── rules/
│   ├── ai_act_fine_guidelines.json      # Tier 4a: Article 99 administrative fine tiers (35M€/7%, 15M€/3%)
│   └── cross_regulatory_frameworks.json # Tier 4b: Multi-framework ontology mappings (NIST, ISO, GDPR)
├── provenance_ledger.json               # Tier 5: Cryptographic Merkle provenance ledger
└── conformity_ground_truth_benchmark.jsonl # Ground-truth evaluation cases

⚖️ Deontic Logic Modal Specification

Every regulatory statement is classified under formal deontic logic:

Deontic Modality	Formal Meaning	Statutory Markers	Example Clause	Fine Exposure
`PROHIBITION`	Forbidden practice; non-compliance is strictly unlawful	"shall not", "prohibited", "unlawful"	Article 5(1)(c) Social Scoring	Up to 35M€ or 7% global turnover
`OBLIGATION`	Mandatory positive duty	"shall", "must", "is required to"	Article 9 Continuous Risk Management	Up to 15M€ or 3% global turnover
`PERMISSION`	Discretionary statutory right	"may", "is entitled to"	Article 10(5) Sensitive data for bias correction	N/A
`EXEMPTION`	Statutory safe harbor or carve-out	"shall not apply to", "derogation"	Article 2(3) Exclusively military / defense AI	Safe Harbor

📊 Dataset Statistics & Coverage

Total Deontic Triples: 36
Total Knowledge Graph Nodes: 105
Total Relational Edges: 222
Articles Grounded: Article 5 (Prohibitions), Article 9 (Risk Management), Article 10 (Data Governance & Bias Mitigation), Article 11 (Annex IV Technical Docs), Article 12 (Automatic Logging), Article 13 (Transparency), Article 14 (Human Oversight & Kill-Switch), Article 15 (Accuracy, Robustness & Cybersecurity), Article 26 (Deployer Duties), Article 27 (FRIA), Article 50 (Generative AI & Deepfakes), Article 51 (GPAI Systemic Risk > 10^25 FLOPs), Article 53 & 55 (GPAI Red-Teaming), Article 99 (Penalties).
Cross-Framework Mappings: 22 bidirectional links to NIST AI RMF 1.0 (GOVERN, MAP, MEASURE, MANAGE), ISO/IEC 42001:2023, and GDPR Articles 22, 25, 32, 35.

💻 Quickstart: Loading in Python

1. Load via Hugging Face `datasets`

from datasets import load_dataset

dataset = load_dataset("gitmodelmujtaba/eu-ai-act-normative-triples", split="train")
print(dataset[0])
# {
#   "id": "EU_AIA_TRIPLE_003",
#   "article_number": 5,
#   "paragraph_number": "1(c)",
#   "modality": "PROHIBITION",
#   "predicate": "shallNotPlaceOnMarketOrPutIntoService",
#   "object": "Social_Scoring_AI_System",
#   "enforcement_tier": "TIER_1_PROHIBITED_AI",
#   ...
# }

2. Load Knowledge Graph into NetworkX

import json
import networkx as nx

with open("data/benchmarks/eu_ai_act_knowledge_graph.json", "r", encoding="utf-8") as f:
    kg = json.load(f)

G = nx.DiGraph()
for node in kg["elements"]["nodes"]:
    G.add_node(node["data"]["id"], **node["data"])

for edge in kg["elements"]["edges"]:
    G.add_edge(edge["data"]["source"], edge["data"]["target"], **edge["data"])

print(f"Graph loaded: {G.number_of_nodes()} nodes, {G.number_of_edges()} edges")

3. Load Cytoscape.js Format in Web UI

The eu_ai_act_knowledge_graph.json contains a direct elements dictionary compatible with Cytoscape.js:

const cy = cytoscape({
  container: document.getElementById('cy'),
  elements: data.elements,
  style: [
    { selector: 'node[category="prohibited_practice"]', style: { 'background-color': '#ef4444', 'label': 'data(label)' } },
    { selector: 'node[category="regulatory_requirement"]', style: { 'background-color': '#3b82f6', 'label': 'data(label)' } },
    { selector: 'edge[modality="PROHIBITION"]', style: { 'line-color': '#ef4444', 'target-arrow-color': '#ef4444', 'target-arrow-shape': 'triangle' } }
  ]
});

🔒 Cryptographic Provenance Ledger

Every file in this benchmark is hashed with SHA-256 and committed to provenance_ledger.json.

Merkle Root Digest: cbf58e52a6f9066ec12826e4fbb7aa267c5a32180566fc0afe47e27eaefe1d08
Genesis Statutory Text: EUR-Lex CELEX:32024R1689 (Official Journal of the European Union, L 2024/1689)
Hash Integrity Guarantee: sha256(source_text) allows zero-hallucination downstream citation auditing.

📑 Citation & BibTeX

@dataset{eu_ai_act_normative_triples_2026,
  author       = {Mujtaba Hussain},
  title        = {EU AI Act Normative Deontic Triples & Knowledge Graph},
  year         = {2026},
  publisher    = {Hugging Face},
  version      = {2.0.0},
  url          = {https://huggingface.co/datasets/gitmodelmujtaba/eu-ai-act-normative-triples}
}

Downloads last month: -

Total file size:

234 kB