Production data should not become public merely because a system needs a demonstration. Customer records, operational identifiers, and commercially sensitive details belong behind the boundary where the software was built to use them.
That does not require replacing the engineering with arbitrary fake data. A useful privacy-safe demonstration can anonymize sensitive values while preserving meaningful relationships: records can still belong to the same groups, sequences can remain coherent, and comparisons can retain the structure the application is designed to investigate.
The distinction is important because fabricated examples often remove the difficult properties that shaped the system. A demo may look complete while no longer exercising realistic variation, traceability, or domain behavior. Anonymization should remove identity, not the relationships needed to evaluate the workflow.
Public AI portfolios have two responsibilities: show credible engineering and respect confidentiality. A strong demonstration does both without implying that private production data is a marketing asset.