# AI models: record what actually changed in a release

https://thiago.limaesilvatecnologia.com.br/en/artigos/model-release-evidence/

Published: 2026-09-12

Behavior depends on more than the model file.

Content produced with AI assistance for Thiago Silva’s website. Independent editorial analysis; it does not represent clients or employers.

Reading-path month: July 2026

Collection published on September 12, 2026. Months organize the reading path; they are not earlier publication dates.

A short architecture article. The technical reference supports the topic; hypothetical scenarios and assessment proposals are editorial analysis, not accounts of personal implementations.

## The architecture decision

An AI update can change data, configuration and inference components. Without a combined record, attributing behavior changes becomes difficult.

## Practical application

Editorial proposal: define a release unit containing relevant dependencies. For a hypothetical classifier, link version, evaluation set and release criteria without unnecessary sensitive data.

## How to verify

Compare product-critical scenarios before promotion and rehearse rollback. Explain differences and limitations. An improved aggregate metric does not finish review when critical cases behave worse.

## Sources

- [OWASP — Secure AI Model Ops Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Secure_AI_Model_Ops_Cheat_Sheet.html). Accessed: 2026-09-12.

[Profile: Thiago Silva | Cyber Architect](https://thiago.limaesilvatecnologia.com.br/en/perfil/)