Synthetic Privacy-Sensitive Text Records with Annotations

Privasis-Zero is a large-scale synthetic dataset containing diverse text records like medical and financial records, legal documents, emails, and messages. It was created by NVIDIA and includes annotations of privacy-related content. The dataset was last updated on March 30, 2026.

Use Cases

• Train models to detect personally identifiable information (PII) using annotated privacy-related content within emails and messages.
• Benchmark text anonymization algorithms on synthetic profile details and surrounding social context from medical and financial records.
• Develop classifiers for document type (e.g., legal vs. medical) based on the textual content of the generated records.
• Generate synthetic training data for privacy-focused NLP tasks using the LLM-generated text records and their annotations.

Strengths

• Dataset is described as large-scale.
• Contains diverse text record types including medical, financial, legal, emails, and messages.
• Includes annotations for privacy-related content.

Limitations

• All data is synthetic, generated by LLMs, which may not perfectly replicate real-world distributions or edge cases.
• Specific row count, column names, and sample size are unknown, making it difficult to assess scale and structure.

Provenance

Source

NVIDIA

Collection Method

Fully generated using LLMs, supplemented with first names from the U.S. Social Security Administration.

Freshness

Last updated on March 30, 2026.

Geography

Primarily U.S. context indicated by the source of first names.