data_validation: key mapping rows on AIBOM Generator field ids

.github/workflows/reference-data-refresh.yml

@@ -28,7 +28,9 @@ jobs:
       - name: Install requirements
         run: python -m pip install --quiet -r data_validation/requirements.txt
 
-      - name: Regenerate ATLAS and CWE tables
+      - name: Regenerate the reference tables
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
         run: python data_validation/reference_data/update_reference_data.py
 
       - name: Run the checks against the new tables
@@ -50,11 +52,11 @@ jobs:
           git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
           git switch -c "$branch"
           git add data_validation/reference_data
-          git commit -m "reference_data: refresh MITRE ATLAS and CWE tables"
+          git commit -m "reference_data: refresh the externally maintained tables"
           git push --force origin "$branch"
-          versions=$(grep -E '^\| (MITRE ATLAS|CWE) ' data_validation/reference_data/SOURCES.md | cut -d'|' -f2,4)
+          versions=$(grep -E '^\| (MITRE ATLAS|CWE|AIBOM Generator fields) ' data_validation/reference_data/SOURCES.md | cut -d'|' -f2,4)
           {
-            echo "The monthly refresh regenerated the MITRE ATLAS and CWE tables."
+            echo "The monthly refresh regenerated the MITRE ATLAS, CWE and AIBOM Generator field tables."
             echo
             echo "Versions: $versions"
             echo

data_validation/reference_data/SOURCES.md

@@ -11,10 +11,11 @@ Lookup tables the validators check identifiers against.
 | Language subtags | `language_subtags.csv` | 2026-09-17 | IANA Language Subtag Registry, `https://www.iana.org/assignments/language-subtag-registry/language-subtag-registry` (languages, scripts, regions) |
 | NIST AI 100-2 | `nist_ai_100_2.csv` | E2025 | NIST AI 100-2e2025, *Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations*, Taxonomy Index, https://doi.org/10.6028/NIST.AI.100-2e2025. The 30 `NISTAML.*` identifiers from its Taxonomy Index and section text (NISTAML.016 and .017 appear only in §3.4.1) |
 | CWE | `cwe_ids.csv` | 4.20 | `https://cwe.mitre.org/data/xml/cwec_latest.xml.zip` (weaknesses, categories and views) |
+| AIBOM Generator fields | `aibom_fields.csv` | aibom-generator@45851cb96dbe | OWASP AIBOM Generator `src/models/field_registry.json`, `https://github.com/GenAI-Security-Project/aibom-generator`: one row per field id, with the CycloneDX path the generator writes it to (`none` when it never writes the field to the BOM), its inclusion rule and its tier |
 
 `owasp_top10.csv` is maintained by hand. The 2026 LLM list renumbered the entries (LLM03 was Supply Chain in 2025 and is Excessive Agency in 2026), so IDs always carry the edition year; `equivalent_2025` gives the 2025 ID each 2026 entry continues (LLM08:2026 Hidden Context Exposure broadens LLM07:2025 System Prompt Leakage). OWASP writes the Agentic IDs without a year (ASI01); this repository adds `:2026` so a future edition cannot collide.
 
-Regenerate the ATLAS, CWE, SPDX and language tables with `python update_reference_data.py`; it fills in the Version column. Do not edit those CSVs by hand. `nist_ai_100_2.csv` and `owasp_top10.csv` are maintained by hand when a new edition is published.
+Regenerate the ATLAS, CWE, SPDX, language and AIBOM field tables with `python update_reference_data.py`; it fills in the Version column. Do not edit those CSVs by hand. `nist_ai_100_2.csv` and `owasp_top10.csv` are maintained by hand when a new edition is published.
 
 `datasets/_shared/dsgai_taxonomy.json` is the DSGAI list the validators use. `dsgai_entries.json` is kept because the Turkish prompt-injection validator reads it.
 

data_validation/reference_data/aibom_fields.csv

@@ -0,0 +1,51 @@
+field_id,cyclonedx_path,rule,tier,description
+attention_head_count,$.components[0].properties,include_if_available,supplementary,Number of attention heads in the model
+attention_head_count_kv,$.components[0].properties,include_if_available,supplementary,Number of Key-Value attention heads
+autonomyType,$.metadata.properties,include_if_available,supplementary,Level of autonomy or human involvement required
+block_count,$.components[0].properties,include_if_available,supplementary,Number of transformer blocks or layers
+bomFormat,$.bomFormat,always_include,critical,Format identifier for the SBOM
+component_version,$.components[0].version,always_include,critical,Version of the component
+context_length,$.components[0].properties,include_if_available,supplementary,Maximum context length or sequence length supported
+datasets,$.component.modelCard.modelParameters.datasets,include_if_available,important,Datasets used for training
+description,$.components[0].description,include_if_available,important,Description of the AI model
+domain,$.metadata.properties,include_if_available,supplementary,Domain or field of application
+downloadLocation,$.component.externalReferences,include_if_available,important,URL to download the model
+embedding_length,$.components[0].properties,include_if_available,supplementary,Vector length of the token embeddings
+energyConsumption,$.metadata.properties,include_if_available,important,Energy consumption information
+energyQuantity,$.metadata.properties,include_if_available,supplementary,Quantitative energy consumption data
+energyUnit,$.metadata.properties,include_if_available,supplementary,Unit of measurement for energy consumption
+ethicalConsiderations,$.component.modelCard.considerations.ethicalConsiderations,include_if_available,important,Ethical considerations and fairness assessments
+external_references,$.component.externalReferences,include_if_available,supplementary,Additional external references
+feed_forward_length,$.components[0].properties,include_if_available,supplementary,Dimensionality of the feed-forward network
+hyperparameter,$.metadata.properties,include_if_available,important,Key hyperparameters of the model architecture
+informationAboutApplication,$.metadata.properties,include_if_available,supplementary,Information about intended applications
+informationAboutTraining,$.metadata.properties,include_if_available,supplementary,Information about the training process
+intendedUse,$.component.modelCard.considerations.useCases,include_if_available,important,Intended use cases for the model
+licenses,$.components[0].licenses,include_if_available,important,License information for the model
+metric,$.metadata.properties,include_if_available,supplementary,Performance metrics and evaluation results
+metricDecisionThreshold,$.metadata.properties,include_if_available,supplementary,Decision thresholds for metrics
+modelDataPreprocessing,$.metadata.properties,include_if_available,supplementary,Data preprocessing information
+modelExplainability,$.metadata.properties,include_if_available,supplementary,Information about model explainability
+model_file_size,$.components[0].properties,include_if_available,supplementary,Total size in bytes of the model weight files in the repository
+name,$.components[0].name,always_include,critical,Name of the AI model component
+paper,none,include_if_present,supplementary,Research paper associated with the model
+primaryPurpose,$.component.modelCard.modelParameters.task,include_if_available,critical,Primary purpose or task of the AI model
+purl,$.components[0].purl,include_if_available,important,Package URL identifier
+quantization_file_type,$.components[0].properties,include_if_available,supplementary,Enum or integer identifier for the quantization bit-precision (e.g. Q4_K_M)
+quantization_version,$.components[0].properties,include_if_available,supplementary,Version or specification identifier of the quantization format
+rope_dimension_count,$.components[0].properties,include_if_available,supplementary,Number of dimensions for Rotary Position Embedding (RoPE)
+runtime_requirement,$.components[0].properties,include_if_available,supplementary,Runtime needed to execute the model based on its weight format: 'python' (loadable from a Python ML stack such as transformers/PyTorch) or 'llm-stack' (requires an LLM inference stack such as llama.cpp/Ollama for GGUF/GGML weights)
+safetyRiskAssessment,$.metadata.properties,include_if_available,important,Safety and risk assessment information
+serialNumber,$.serialNumber,always_include,critical,Unique identifier for this SBOM instance
+specVersion,$.specVersion,always_include,critical,CycloneDX specification version
+standardCompliance,$.metadata.properties,include_if_available,supplementary,Standards or regulations the model complies with
+suppliedBy,$.component.supplier,include_if_available,critical,Organization or individual that supplied the model
+technicalLimitations,$.component.modelCard.considerations.technicalLimitations,include_if_available,important,Known limitations of the model
+tokenizer_class,$.components[0].properties,include_if_available,supplementary,The specific tokenizer class or method used
+type,$.components[0].type,always_include,critical,Type of component (machine-learning-model)
+typeOfModel,$.components[0].modelCard.modelParameters.modelArchitecture,include_if_available,important,Type or architecture of the model
+useSensitivePersonalInformation,$.metadata.properties,include_if_available,supplementary,Information about use of sensitive personal data
+vcs,none,include_if_present,supplementary,Version control system URL
+version,$.version,always_include,critical,Version of this SBOM document
+vocab_size,$.components[0].properties,include_if_available,supplementary,Expected size of the model's vocabulary
+website,none,include_if_present,supplementary,Model website or documentation URL

data_validation/reference_data/update_reference_data.py

@@ -1,12 +1,14 @@
 """
 Regenerates the lookup tables that track an external release: MITRE ATLAS,
-CWE, the SPDX License List and the IANA Language Subtag Registry.
+CWE, the SPDX License List, the IANA Language Subtag Registry and the OWASP
+AIBOM Generator field registry.
 
     python update_reference_data.py              # all of them
     python update_reference_data.py --atlas      # mitre_atlas_techniques.csv
     python update_reference_data.py --cwe        # cwe_ids.csv
     python update_reference_data.py --spdx       # spdx_licenses.csv
     python update_reference_data.py --languages  # language_subtags.csv
+    python update_reference_data.py --aibom      # aibom_fields.csv
 
 Run it when one of them publishes a new release (the reference-data-refresh
 workflow does so monthly), and commit the regenerated CSVs together with the
@@ -20,6 +22,7 @@
 import csv
 import io
 import json
+import os
 import re
 import sys
 import urllib.request
@@ -42,11 +45,18 @@
 CWE_URL = "https://cwe.mitre.org/data/xml/cwec_latest.xml.zip"
 SPDX_URL = "https://raw.githubusercontent.com/spdx/license-list-data/main/json/licenses.json"
 IANA_SUBTAGS_URL = "https://www.iana.org/assignments/language-subtag-registry/language-subtag-registry"
+AIBOM_REPO = "GenAI-Security-Project/aibom-generator"
+AIBOM_HEAD_URL = f"https://api.github.com/repos/{AIBOM_REPO}/commits/main"
+AIBOM_REGISTRY_URL = f"https://raw.githubusercontent.com/{AIBOM_REPO}/%s/src/models/field_registry.json"
 SOURCES = HERE / "SOURCES.md"
 
 
 def fetch(url: str) -> bytes:
-    with urllib.request.urlopen(url, timeout=60) as resp:  # noqa: S310 - fixed https URLs
+    request = urllib.request.Request(url)  # noqa: S310 - fixed https URLs
+    token = os.environ.get("GITHUB_TOKEN")
+    if token and url.startswith("https://api.github.com/"):
+        request.add_header("Authorization", f"Bearer {token}")
+    with urllib.request.urlopen(request, timeout=60) as resp:  # noqa: S310 - fixed https URLs
         return resp.read()
 
 
@@ -156,13 +166,28 @@ def update_languages() -> None:
     print(f"language_subtags.csv: {len(rows)} subtags (IANA registry {file_date})")
 
 
+def update_aibom() -> None:
+    """The fields the OWASP AIBOM Generator can write, from its field_registry.json at the current main commit."""
+    commit = json.loads(fetch(AIBOM_HEAD_URL))["sha"]
+    fields = json.loads(fetch(AIBOM_REGISTRY_URL % commit))["fields"]
+    rows = [{"field_id": fid, "cyclonedx_path": f["aibom_generation"]["location"], "rule": f["aibom_generation"]["rule"],
+             "tier": f["tier"], "description": f["description"]} for fid, f in sorted(fields.items())]
+    with open(HERE / "aibom_fields.csv", "w", newline="", encoding="utf-8") as fh:
+        w = csv.DictWriter(fh, fieldnames=["field_id", "cyclonedx_path", "rule", "tier", "description"], lineterminator="\n")
+        w.writeheader()
+        w.writerows(rows)
+    set_version("AIBOM Generator fields", f"aibom-generator@{commit[:12]}")
+    print(f"aibom_fields.csv: {len(rows)} fields (aibom-generator {commit[:12]})")
+
+
 def main(argv: list[str] | None = None) -> int:
     parser = argparse.ArgumentParser(description="Regenerate the externally maintained reference tables")
-    for name in ("atlas", "cwe", "spdx", "languages"):
+    for name in ("atlas", "cwe", "spdx", "languages", "aibom"):
         parser.add_argument(f"--{name}", action="store_true")
     args = parser.parse_args(argv)
-    every = not (args.atlas or args.cwe or args.spdx or args.languages)
-    for name, update in (("atlas", update_atlas), ("cwe", update_cwe), ("spdx", update_spdx), ("languages", update_languages)):
+    every = not (args.atlas or args.cwe or args.spdx or args.languages or args.aibom)
+    for name, update in (("atlas", update_atlas), ("cwe", update_cwe), ("spdx", update_spdx), ("languages", update_languages),
+                         ("aibom", update_aibom)):
         if every or getattr(args, name):
             update()
     return 0

data_validation/requirements-dev.txt

@@ -1,6 +1,10 @@
 # Hash-pinned lock generated from requirements-dev.in by data_validation/update_locks.py
 # (uv pip compile --universal --generate-hashes --python-version 3.11).
 # Do not edit by hand: change requirements-dev.in and rerun the script.
+agent-evidence-vectors==0.24.0 \
+    --hash=sha256:bc8c5785dc9768676361151a4ca2fb4dc1d9bd5a22a67381f6db7adda014539d \
+    --hash=sha256:fa1c4b804f700a0f71f1a1776978546d9df571b4e5a451b58220478f75e3b6d3
+    # via -r requirements.in
 arrow==1.4.0 \
     --hash=sha256:749f0769958ebdc79c173ff0b0670d59051a535fa26e8eba02953dc19eb43205 \
     --hash=sha256:ed0cc050e98001b8779e84d461b0098c4ac597e88704a655582b21d116e526d7

data_validation/requirements.in

@@ -5,3 +5,7 @@ jsonschema[format-nongpl]>=4.18.0
 pytest>=7.0.0
 # Only for reference_data/update_reference_data.py (refreshing the ATLAS table)
 pyyaml>=6.0
+# Signed conformance records and their reference verifier, replayed by
+# validators/evidence_mapping_check.py for mapping rows that cite test_records.
+# Exact pin: a new corpus release can change a verdict, so Dependabot proposes each bump.
+agent-evidence-vectors==0.24.0

data_validation/requirements.txt

@@ -1,6 +1,10 @@
 # Hash-pinned lock generated from requirements.in by data_validation/update_locks.py
 # (uv pip compile --universal --generate-hashes --python-version 3.11).
 # Do not edit by hand: change requirements.in and rerun the script.
+agent-evidence-vectors==0.24.0 \
+    --hash=sha256:bc8c5785dc9768676361151a4ca2fb4dc1d9bd5a22a67381f6db7adda014539d \
+    --hash=sha256:fa1c4b804f700a0f71f1a1776978546d9df571b4e5a451b58220478f75e3b6d3
+    # via -r requirements.in
 arrow==1.4.0 \
     --hash=sha256:749f0769958ebdc79c173ff0b0670d59051a535fa26e8eba02953dc19eb43205 \
     --hash=sha256:ed0cc050e98001b8779e84d461b0098c4ac597e88704a655582b21d116e526d7

data_validation/run_all_checks.py

@@ -10,8 +10,9 @@
 2. Shared checks (data_validation/validators/) on every data file:
    schema (for files no dataset validator covers), DSGAI IDs, CVE/CWE/ATLAS
    OWASP Top 10 and NIST AI 100-2 IDs, CVSS scores and ratings, dates,
-   SPDX licences and BCP 47 language tags, anonymization patterns and
-   duplicates. Errors fail the run;
+   SPDX licences and BCP 47 language tags, mapping rows that cite
+   published test records, mapping rows keyed on AIBOM Generator field ids,
+   anonymization patterns and duplicates. Errors fail the run;
    heuristic warnings (anonymization, near-duplicates, retired IDs) are
    counted, and printed with --verbose.
 
@@ -36,11 +37,13 @@
 sys.path.insert(0, str(Path(__file__).resolve().parent))
 
 from validators import (  # noqa: E402
+    aibom_field_check,
     anonymization_scanner,
     crossref_validator,
     date_check,
     dedup_checker,
     dsgai_mapping_check,
+    evidence_mapping_check,
     metadata_check,
     schema_validator,
     severity_check,
@@ -158,6 +161,8 @@ def shared_checks(dataset: Path, datasets_root: Path) -> tuple[int, list[Finding
         findings += severity_check.check_data(path, data)
         findings += date_check.check_data(path, data)
         findings += metadata_check.check_data(path, data)
+        findings += evidence_mapping_check.check_data(path, data)
+        findings += aibom_field_check.check_data(path, data)
         findings += anonymization_scanner.check_data(path, data)
     records, load_errors = dedup_checker.load_records(files)
     findings += load_errors + dedup_checker.check_records(records)
@@ -225,7 +230,10 @@ def main(argv: list[str] | None = None) -> int:
 
     checked_files = errors = warnings = 0
     if not args.no_shared:
-        print("\nShared checks (schema, DSGAI, CVE/CWE/ATLAS/OWASP/NIST IDs, CVSS, dates, licences, languages, anonymization, duplicates):")
+        print(
+            "\nShared checks (schema, DSGAI, CVE/CWE/ATLAS/OWASP/NIST IDs, CVSS, dates, licences, languages,"
+            " test records, anonymization, duplicates):"
+        )
         for dataset in datasets:
             n, findings = shared_checks(dataset, datasets_root)
             if not n: