Add data attribution section for AI transparency
Hello OpenAI team, and congratulations on the Privacy Filter release.
The current model card does not list the datasets used for training or evaluation. In the interest of AI transparency and data provenance, this PR proposes an Acknowledgements and Data Attribution section with a suggested citation. If Ai4Privacy data was not used for this model, we would still welcome a short training-data disclosure section, and are happy to adjust or close this PR accordingly.
Update: I have revised this PR. The release announcement already credits Ai4Privacy, reporting evaluation on our PII-Masking-300k benchmark (F1 96%, and 97.43% on a corrected version). The proposed section now simply mirrors that existing attribution on the model card itself, so the evaluation provenance is visible where the model is distributed. Thank you for the citation in the announcement, and congratulations again on the release.
Simplified the PR to a metadata-only change: it now just adds ai4privacy/pii-masking-300k to the datasets field of the model card frontmatter, matching the evaluation reported in the release announcement. No body text is modified. This links the model to the benchmark on the Hub for transparency and discoverability.