Masked translation
PII and personal information masking before AI translation
Many teams want AI translation quality, but they do not want customer names, IDs, phones, emails, bank accounts or company names to leave with the body text. In search terms, that problem may be called PII masking, Personal Information protection, or document anonymization before translation: replace sensitive entities with stable placeholders first.
Veil Translate handles PII detection, review, masking and restoration in the browser. The outbound translation request contains masked text, not the original sensitive values.
Short answer
PII masking before AI translation means replacing confirmed personal information and sensitive values with stable placeholders in the browser before any translation request is sent. The model sees the placeholder version, preserves those tokens, and the browser restores the real values locally after translation.
What PII masking before AI translation means
It does not mean sending an encrypted document to the model. It means replacing identifiable sensitive entities with readable placeholders such as ⟦EMAIL_1⟧, ⟦ID_1⟧ and ⟦BANK_1⟧.
The model translates normal body text and preserves placeholders. After translation returns, the browser restores placeholders from the local mapping table.
Documents this workflow fits
Use it for contracts, NDAs, resumes, candidate packets, quotes, customer lists, financial spreadsheets and cross-border business materials.
It is especially useful when the translated file must keep the Word, Excel, PowerPoint or text-based PDF layout instead of becoming a pasted chat transcript.
Example masked segment
Candidate Zhao Lei uses zhaolei@example.com and ID number 110101199003070012.
Candidate ⟦NAME_1⟧ uses ⟦EMAIL_1⟧ and ID number ⟦ID_1⟧.
候选人⟦NAME_1⟧的邮箱是⟦EMAIL_1⟧,身份证号为⟦ID_1⟧。
Privacy boundary
This is not encrypted translation. Confirmed sensitive entities are replaced with placeholders; the remaining de-identified body text is still sent as plaintext to the provider you choose.
Automatic detection is strongest for Chinese and English source text. For other source languages, manually add the values you want protected on the review screen.
Frequently asked questions
Is PII masking the same as document anonymization?
In this workflow, both mean replacing sensitive entities with placeholders. Translation mode keeps a local mapping for restoration; anonymize-only mode creates a one-way masked copy.
Will placeholders hurt translation quality?
Usually not for the main meaning. Placeholders preserve entity positions and are restored locally after the translated text returns.
Are glossary terms sent out?
Glossary matches can be masked as TERM placeholders, so the configured source value does not need to be sent directly in the translation payload.
Do I need my own API key?
For BYO-key translation, yes. The browser sends the request directly to the OpenAI-compatible endpoint you choose.