OrcaRouter Adds Data Cloak Feature That Swaps Personal Data for Synthetic Values
The design target is specific: keep personal information out of the model's input without the task-quality drop that comes from masking values outright, which is the tradeoff that has kept PII-heavy workflows off hosted models.
Reporting from 1 source: ASCII.jp.
FlashLabs Inc. has begun offering Data Cloak in OrcaRouter, the AI inference gateway developed by Continuum AI and provided by FlashLabs for the Japanese market. When enabled, names, email addresses, phone numbers and other identifying information in a request are replaced with synthetic values before reaching the model, then restored automatically in the response. FlashLabs says the feature is free to all users.
Data Cloak is switched on from a Guardrails screen and needs no application code changes, according to FlashLabs. The replaced values are realistic, not tags: an email address is swapped for a different address in a usable format, so the model can still draft a message or format an address. Secret material such as API keys, private keys and tokens is not replaced but blocked from being sent. The mapping between real and synthetic values is held only in memory and is not written to disk.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.