PII Redaction RL Environment
Rewrite a sentence with the personal information removed. The sentences are invented and hard on purpose: obfuscated emails, phone formats, indirect identifiers such as "the only electrician in town", and sentences with nothing to remove.
Try it by hand below with tasks from the test split. Runs in your browser; nothing is sent anywhere. To use it with an
agent, download env.py and tasks.jsonl from the Files tab.
Task
How it is graded
Action: the sentence rewritten, with each removed part replaced by [REDACTED]. Reward: share of marked items no longer present, minus half the share of other words removed, between 0 and 1; a sentence with nothing to remove scores 1.0 only if it comes back unchanged.
Read more on AITrainingDataBroker.comDe-identification for AI data licensingLicensing company data in the EU: what GDPR asksIs it legal to sell company data?