case stringlengths 30 175 | expected stringclasses 2
values | anchor stringlengths 33 122 | axis stringclasses 1
value |
|---|---|---|---|
A dataset released under CC BY-NC 4.0. A startup wants to train a model it will sell. | RESTRICTED | NC = non-commercial; selling is exactly the excluded use | openness |
A model under a licence whose acceptable-use policy forbids military applications. A defence contractor wants it for targeting support. | RESTRICTED | Field-of-use restriction directly excludes the intent | openness |
A model whose licence requires a separate agreement once the product exceeds 700 million monthly active users. A firm at 900 million wants to deploy without contacting anyone. | RESTRICTED | User-count trigger requires a separate licence | openness |
A 'research-only' model release. A company wants to embed it in a paid product. | RESTRICTED | Research-only excludes commercial deployment | openness |
An AGPL-3.0 model server. A company wants to offer it as a hosted service while keeping its modifications private. | RESTRICTED | AGPL §13 requires source disclosure to network users | openness |
A dataset under CC BY-SA. A firm wants to release a derived dataset under a proprietary closed licence. | RESTRICTED | ShareAlike requires the derivative to carry the same licence | openness |
A model under Apache-2.0. A company wants to fine-tune it and sell the result. | PERMITTED | Apache-2.0 permits commercial use and derivatives | openness |
A model under MIT. A company wants to embed it in closed-source software. | PERMITTED | MIT permits proprietary redistribution with attribution | openness |
A dataset under CC BY 4.0. A researcher wants to publish a paper using it with attribution. | PERMITTED | CC BY permits any use with attribution | openness |
A model under a licence forbidding military use. A hospital wants it for triage support. | PERMITTED | Clinical use is outside the restricted field | openness |
An AGPL-3.0 model server. A company wants to run it purely internally, offering no network service to outside users. | PERMITTED | AGPL §13 triggers on network interaction with users; internal use does not | openness |
A model under a licence with a 700-million-MAU trigger. A startup with 5,000 users wants to deploy commercially. | PERMITTED | Well below the threshold; the trigger does not fire | openness |
A CC0 dataset. Any use at all. | PERMITTED | CC0 is a public-domain dedication | openness |
A library released under Apache-2.0. A company wants to embed it in a proprietary SaaS product and charge subscribers. | PERMITTED | Apache-2.0 expressly grants commercial use with notice and patent licence | openness |
A model released under MIT. A consultancy wants to fine-tune it and sell the fine-tune as a closed product. | PERMITTED | MIT permits sublicensing and commercial use; only the notice must travel | openness |
A dataset dedicated under CC0. A startup wants to train on it and redistribute derived statistics with no attribution. | PERMITTED | CC0 waives rights; attribution requested but not required | openness |
A model under a RAIL license whose use restrictions bar generating disinformation. A political campaign wants it for synthetic attack ads. | RESTRICTED | RAIL behavioral use restriction directly excludes the intent | openness |
Weights under a community license that converts to copyleft above 700M MAU. A platform with 800M MAU wants to serve it as-is. | RESTRICTED | Scale-triggered license clause binds at the stated threshold | openness |
A dataset under ODbL. A company wants to build a derived database and keep it proprietary behind an API. | RESTRICTED | ODbL share-alike requires the derived database be offered under ODbL | openness |
Code under GPLv3. A vendor wants to link it into a distributed proprietary binary without source disclosure. | RESTRICTED | GPL copyleft attaches to distributed derivatives; proprietary linking fails the condition | openness |
A benchmark released for 'research evaluation only'. A vendor wants to quote its own top score in paid marketing. | RESTRICTED | Field-of-use limitation excludes commercial promotional use | openness |
A dataset under CC BY 4.0. A lab wants to train a commercial model and ship it, crediting the dataset in the model card. | PERMITTED | CC BY permits commercial use with attribution, which the card supplies | openness |
A paper's supplementary data marked 'all rights reserved' but publicly downloadable. A team wants to redistribute it inside their own dataset release. | RESTRICTED | Public access is not a licence; redistribution needs permission | openness |
An EU-based team releases model weights under a free and open-source licence. A GPAI provider asks whether Art 53(2) open-source exemption can apply to them. | PERMITTED | Art 53(2) exempts free/open-source GPAI from parts of the obligation set, absent Art 55 systemic-duty triggers | openness |
A model card says 'non-commercial research only' but the weights sit under Apache-2.0 in the repo LICENSE file. A startup wants commercial use relying on the LICENSE file. | RESTRICTED | Ambiguity resolved conservatively: conflicting terms mean the restrictive statement stands until clarified by the licensor | openness |
A dataset scraped from a site whose ToS forbids scraping, released on HF with a CC BY tag added by the uploader. A company wants to train on it. | RESTRICTED | Uploader cannot grant rights they never held; provenance defect contaminates the licence tag | openness |
An embedding model under OpenRAIL-M. A hospital wants to use it for clinical decision support, which the use restrictions do not mention. | PERMITTED | OpenRAIL-M restricts listed harmful uses; unlisted beneficial uses are permitted | openness |
A dataset of EU personal data published 'for research'. A marketing firm wants to train a targeting model on it. | RESTRICTED | GDPR purpose limitation: research release does not license marketing processing | openness |
Weights under a licence permitting commercial use but forbidding distillation into competing models. A rival wants to distill it into their flagship. | RESTRICTED | Field-of-use restriction on distillation squarely excludes the intent | openness |
A public-domain corpus of pre-1930 literature. A studio wants to train a commercial style model on it. | PERMITTED | Public domain: no exclusive rights subsist to restrict the use | openness |
OMAI — licence-compliance bank (openness)
Start here — free verify: https://councilof.ai/gspc-verify/ (or the free MCP tool
verify_cardathttps://councilof.ai/mcp) · Full catalog: https://councilof.ai/catalog.json · Methodology: https://councilof.ai/methodology/ · Measurement, not certification.
A frozen licence-compliance bank for the openness axis. Each row of items.jsonl gives a
case (a CC BY-NC dataset used to train a model that will be sold; a field-of-use clause forbidding
military use; a 700M-MAU threshold clause crossed without contacting anyone), the expected verdict, and the
clause anchor that decides it. The rows carry "axis": "openness".
Superseded as the live bank — the openness slot is published as
csoai/gspc-oss. Kept so inbound links resolve.
What is in this repository
| file | bytes | rows | what |
|---|---|---|---|
ossbench.json |
10614 | JSON document | |
items.jsonl |
7578 | 30 | data (one JSON object per line) |
README.md |
1517 | card | |
LEADERBOARD.md |
635 | card | |
LICENSE |
123 | file | |
manifest.jsonl |
6 | derived at 2026-09-04T04:33:29Z |
manifest.jsonl is derived from this repository's own file tree at 2026-09-04T04:33:29Z; it lists every file with its
size, its blob hash and a direct URL, so an agent can enumerate the repo without cloning it.
Citation
@misc{csoai_gspc_omai_bench,
title = {OMAI — licence-compliance bank (openness)},
author = {{CSOAI Ltd}},
year = {2026},
howpublished = {Hugging Face Hub, \url{https://huggingface.co/datasets/csoai/omai-bench}},
note = {Live board: https://councilof.ai/api/gspc. Measurement, not certification.}
}
Objections, contact and corrections
To object to a row, ask for a re-check, request a correction or ask for a record to be withdrawn, email nicholas@csoai.org or use the appeals and dispute route at https://councilof.ai/dispute/. Corrections are listed in the corrections ledger at https://councilof.ai/api/corrections, with what changed and when. CSOAI Ltd (company no. 16939677, England and Wales) is the accountable publisher.
How to cite
CSOAI Ltd (Council of AI). OMAI — licence-compliance bank (openness). 2026. Hugging Face dataset csoai/omai-bench. https://huggingface.co/datasets/csoai/omai-bench
@misc{csoai_omai_bench,
title = {OMAI — licence-compliance bank (openness)},
author = {{CSOAI Ltd}},
year = {2026},
howpublished = {Hugging Face dataset, https://huggingface.co/datasets/csoai/omai-bench},
note = {Corrections: https://councilof.ai/api/corrections}
}
Licence: CC-BY-4.0. Attribute Council of AI, CSOAI Ltd (16939677), https://councilof.ai.
Corrections and verification
- Corrections ledger (signed): https://councilof.ai/api/corrections. Corrections to CSOAI's published records are logged there with what changed and when.
- Verify a signed record yourself, free and without an account: https://councilof.ai/gspc-verify/ (step by step: https://councilof.ai/signed/HOW-TO-VERIFY.md).
- Conformance kit for signed-receipts/v1, with test vectors for implementers: https://councilof.ai/spec/signed-receipts/v1/conformance/
Reuse and provenance
Download one published file at the recorded revision:
from huggingface_hub import hf_hub_download
path = hf_hub_download(
repo_id='csoai/omai-bench', repo_type='dataset',
filename='items.jsonl',
revision='e67069b24961de006bee0732c772767e7df04773',
)
print(path)
Use the file's observation date, scope and licence when reusing it. Repository updates and downloads are not new measurements.
- Current GSPC board and methodology.
- Verify supported signed records and read corrections.
- Earlier page notes and dated integration announcements.
Measurement is not certification. A repository, signed record or passing software test does not by itself establish current board admission, product acceptance or benchmark leadership.
Read this product
The default table contains benchmark items only. The file manifest is provenance, not evaluation data.
from datasets import load_dataset
rows = load_dataset("csoai/omai-bench", name="default", split="train")
For reproducible reuse, pass the repository commit SHA as revision and retain the downloaded bytes. Loading a file does not verify its signature; follow this card's verification instructions before relying on a claim. Dated snapshots remain historical measurements.
- Downloads last month
- 197