# Air Datasets

> Air Datasets publishes AI4PRIVACY, the largest open datasets of realistic, entirely artificial (synthetic) personal data, built for training and evaluating models that find and mask personal information (PII) in text and speech. The datasets have been downloaded over 15 million times and are used in published research worldwide. Air Datasets is an Ai4Privacy initiative of AI Suisse SA, built in Switzerland.

Synthetic personal data means no real identity is ever at risk: teams can train, mask and evaluate privacy models, share the data across teams and borders, and license it clearly. When research needs real data for analytics, Air Datasets connects teams with real-data partners.

## Pages

- [Home](https://airdatasets.com/): What Air Datasets is and why synthetic personal data removes the privacy obstacle in research.
- [AI4PRIVACY datasets](https://airdatasets.com/ai4privacy): The dataset catalogue (OpenPII 1.5M, OpenPII 1M, PII-Masking-3M, 400k, 300k, 200k, 65k, 43k; PII-Masking-5M text and audio in preparation) and a benchmark: a model trained on these datasets reaches 0.92 F1 on a held-out test set versus 0.30 off the shelf (internal evaluation, July 2026).
- [The Air Data Gazette](https://airdatasets.com/gazette): Register of external publications that used the datasets, from Google DeepMind to TU Wien.
- [Why synthetic data](https://airdatasets.com/mission): Where synthetic data is better than real (training), strong (benchmarking, prototyping) and where to escalate to real data (analytics).

## Related

- [Ai4Privacy](https://www.ai4privacy.com/): The privacy project behind the datasets: open datasets, privacy-preserving AI chat, PII detection API and local SDKs.
- [Ai4Privacy llms.txt](https://www.ai4privacy.com/llms.txt): Full Ai4Privacy product and dataset index.
- [Hugging Face: ai4privacy](https://huggingface.co/ai4privacy): Download the open datasets and models.

## Contact

- Licensing and access: hello@aisuisse.com