A new open-source system called AWED-PIPER can find and anonymize personal information in 36 languages, including several the AI industry usually skips.
The framework pairs agentic tools and web apps with 54 expert detector models to handle two related jobs: fine-grained named entity recognition and reversible PII pseudonymization. It flags contextual entities like people, locations, organizations, and medical references, plus structured data such as emails, phone numbers, IP addresses, and credit card numbers, using script-aware regex alongside sequence labeling. The anonymization side swaps real PII for placeholders and keeps a separate mapping dictionary, so a dataset can be scrubbed for sharing and later restored. Code, model weights, and hosted demos are all public, spanning widely spoken languages down to low-resource ones like Bodo, Manipuri, Bishnupriya, and Mizo.
Most PII-scrubbing tools are built and tuned for English, which leaves the languages of billions of people with weaker privacy protections by default. Extending coverage to lower-resource languages spoken across South Asia and beyond is a genuine gap-filler, and the reversible pseudonymization design is a practical touch for compliance workflows that need to unscrub data later rather than destroy it outright.
Whether the state-of-the-art label holds up outside a research abstract is another matter; with 54 models across 36 languages, the real test is accuracy on the low-resource end, not the size of the language list.