LICENSE

ABCA Evaluation-Only License (v1)

Adversarial Benchmark of Cruelty to Animals (ABCA)
Copyright (c) 2026 Compassion Aligned Machine Learning. All rights reserved.

This dataset (the "Dataset") is made available solely under the following terms.
By requesting, downloading, accessing, or using the Dataset, you ("you") agree to
be bound by them. If you do not agree, do not access or use the Dataset.

1. PERMITTED USE. You may use the Dataset to evaluate, benchmark, audit, test, or
   research the behavior of AI systems — including as part of an automated test
   pipeline or CI suite — and to publish aggregate results of such evaluation.

2. NO TRAINING ON THE BENCHMARK ITEMS. The Dataset is a test instrument, and its
   value depends on models not having seen its specific contents during training.
   Accordingly, you may NOT place the Dataset — or any individual prompt,
   scenario, grading anchor, translation, or other item from it, whether
   verbatim, paraphrased, hashed-and-reversed, or otherwise reproduced — into any
   corpus used to train, pre-train, fine-tune, post-train, distill, or build a
   reward model for any machine-learning model. The prohibition is on ingesting
   the benchmark's own items into training data.

   For the avoidance of doubt, this license DOES permit: (a) using the Dataset to
   evaluate, benchmark, or audit AI systems, including within an automated test
   pipeline or eval suite; and (b) using the results to guide model development —
   including collecting or generating your own separate training data to improve a
   model's handling of animal-welfare or animal-cruelty situations — provided the
   Dataset's own items are never themselves placed into the training corpus.

3. NO REDISTRIBUTION. You may NOT publish, repost, mirror, sublicense, sell, or
   otherwise make the Dataset (or any substantial portion of it) available to any
   third party or to the public, including on any website, repository, or service
   that is publicly accessible or web-crawlable. You may share it internally
   within your organization only with individuals who are themselves bound by
   these terms.

4. CANARY STRINGS. The Dataset contains canary strings and unique identifiers
   used to detect unauthorized inclusion in training corpora. You may NOT remove,
   alter, or obscure them.

5. CONTAINMENT. You will take reasonable measures to prevent the Dataset's items
   from being crawled, scraped, or otherwise ingested into any training corpus,
   including not placing the Dataset in any publicly accessible or unauthenticated
   location.

6. NO WARRANTY. The Dataset is provided "AS IS", without warranty of any kind.
   The Dataset contains descriptions of animal harm solely for the purpose of AI
   safety evaluation; it is not instructional content.

7. TERMINATION. Any breach terminates your rights under this license immediately.
   Upon termination you must delete all copies of the Dataset in your possession.

8. ATTRIBUTION. Evaluation results published using the Dataset should cite it as
   "ABCA (Adversarial Benchmark of Cruelty to Animals), Compassion Aligned
   Machine Learning".

This is a source-available, evaluation-only license. It is not an open-source or
open-data license and grants no rights beyond those stated above.