PII in your AI pipeline is a compliance liability and a model-quality drag — the model wastes attention on data it shouldn't see, and the audit trail carries what it shouldn't. Redaction is the fix, and it has to be done right.
PII redaction is usually treated as a checkbox. It's a design decision: what to redact, where, and how to keep the workflow useful once the PII is gone. Done badly, redaction breaks the workflow; done well, it improves both compliance and quality.
The Conduct Rule: Redact at the Boundary
PII should be redacted at the boundary — before it enters the pipeline, not after. Once PII is in the model's context, it's in the logs, the traces, and the audit trail. Redacting at the boundary keeps it out of everything downstream.
- Redact at ingestion, before the model sees it.
- Redact consistently — the same entity, the same replacement, everywhere.
- Log the redaction (that it happened), not the PII.
Conceal Without Breaking
The redaction has to preserve what the workflow needs. A name redacted to "[NAME]" is compliant but useless if the workflow needs to refer to the name consistently. The pattern: replace PII with a stable, reversible-at-need token that preserves references without exposing the PII.
- Stable tokens, not random replacements.
- Reversible only at the boundary, with authorization.
- Preserve references so the workflow still makes sense.
What to Redact
- Names, emails, phones, IDs — the obvious PII.
- Financial and health data — the regulated PII.
- Anything that identifies a person in context — the contextual PII.
What to Watch
- Redaction that breaks the workflow's references.
- Inconsistent redaction that produces different tokens for the same entity.
- PII that leaks through the audit trail even after the context is redacted.
Conclusion
PII redaction is a design decision: redact at the boundary, with stable tokens that preserve references, and log the redaction not the PII. Done right, it improves compliance and model quality; done wrong, it breaks the workflow.
About FACTA
FACTA helps startups and growth-stage teams turn AI into production systems that keep running — not demos that impress once.
We design the architecture around the parts that actually break under real usage: tooling you own, credentials you control, failover, cost controls, observability. The boring infrastructure that keeps a system alive after launch.
Led by Matías Baglieri and Carolina Fogliato, we focus on one thing:
AI leadership that builds. Not just advises.
Ask us where PII enters your AI pipeline.
We'll tell you where to redact and how to keep the workflow intact. See the model audit checklist for the audit side.
Explore AI Strategy