← Index

SCORESecurity & safety

RelayClear POC for Identity Anonymization

Jonas@jonasplouffe
Development screenshot of RelayClear using fictional test data. A sample workplace feedback submission is shown on the left. The system highlights details that may identify the submitter, including a direct self-identification, “My name is Steve Johns.” On the right, the assessment returns High internal risk with 99% confidence and attribution to the submitter.Development screenshot of RelayClear using fictional test data. A sample workplace feedback submission is shown on the left. The system highlights contextual details including specific locations, dates, and times that could narrow down who submitted the feedback. On the right, the assessment returns Medium internal risk with 56% confidence and shared attribution.

This project built a proof-of-concept (POC) to determine if free-text feedback could reveal the identity of the submitter, utilizing the Jev model via Cloudflare Workers AI.

Built a POC this morning for RelayClear using @typesafeai's Jev model via @Cloudflare Workers AI to assess whether free-text feedback could reveal the identity of the submitter. Surprised me how well it understands the distinction between identifying someone mentioned in the feedback vs identifying the person submitting it. This did not take long to build. Still a POC, but feels promising.
Original post