agentlans/personal-information-prompts

Personal Information Prompts

This dataset contains multilingual prompts derived from the all_sample subset of the agentlans/allenai-WildChat-4.8M dataset. Each prompt features artificially inserted personally identifiable information (PII) generated randomly with the Faker Python package for various locales.

Each rewritten prompt uses the google/gemma-3-12b-it model to incorporate the synthetic personal data.

Dataset fields for the two configurations:

classification configuration

This config was created from the customized_prompts config and then split into 80% training and 20% testing data.

• —prompt: a prompt which may or may not contain PII
• —language: the detected language of the prompt
• —contains_pii: whether PII was used in constructing the prompt

customized_prompts configuration

• —prompt: prompt text including artificial personal data
• —pii: dictionary containing the PII used for generation
• —base_prompt: original prompt from the WildChat dataset
• —pii_locale: locale used to generate the PII
• —prompt_language: language of the rewritten prompt, determined with FastText

Limitations

• —All information randomly generated except for the base prompts. Any resemblance to actual events or locales or persons, living or dead, is entirely coincidental.
• —This dataset is not suitable for extracting PII because parts of the PII get embedded into foreign language texts in unusual ways.
• —Not all the PII gets incorporated into the prompt.

Licence

Creative Commons Attribution 4.0