# Shadow Deployment

> A new model gets copies of real requests. Its answers are saved but not shown.

**Field:** Artificial Intelligence  
**Category:** MLOps & Deployment  
**Also known as:** Shadow Mode, Shadow Testing, Dark Launch (ML)

## Think of it like

It is like a new driver who sits in the car and says what they would do. The real driver still steers.

## Definition

Shadow deployment is a way to test a new model on real traffic without users seeing its answers. Each live request still goes to the old model, which answers as usual. A copy of the request also goes to the new model. Its answers are saved and compared, but never shown. If the new model does well, it can replace the old one with less risk.

## Example

Lucy's bank has a new fraud model. It runs in shadow for two weeks, and the team checks its flags against the old model's before switching.

## Why it matters

Test data never matches real life. Shadow mode tests a model on real traffic with no risk to users.

## Common mistake

Letting the shadow model take actions. If it sends emails or blocks cards, it is not a shadow any more.

## Three principles

- **Real traffic.** The new model sees live requests.
- **Hidden answers.** Saved, compared, never shown.
- **Switch with proof.** Swap in only after it does well.

## Related terms

- [Model Serving](https://learn.benibix.com/term/ai/model-serving)
- [Canary Release](https://learn.benibix.com/term/development/canary-release)
- [A/B Test](https://learn.benibix.com/term/marketing/ab-test)
- [Model Monitoring](https://learn.benibix.com/term/ai/model-monitoring)
- [Model Registry](https://learn.benibix.com/term/ai/model-registry)

## Told apart from

- [Shadow Deployment vs Canary Release](https://learn.benibix.com/compare/shadow-deployment-vs-canary-release): People think a shadow deployment and a canary release are the same careful way to launch a model, and they are not. In a canary, a few real users get the new version and see what it does. In a shadow deployment, no user sees the new model at all. It gets a copy of real requests, and its answers are only saved. A canary tests on people. A shadow tests in secret.

![Four panels. A model trained in PyTorch is saved as one ONNX file and runs on a phone, a browser and a server. A big slow model becomes a small fast one. Live requests go to the old model and a copy goes to a new model whose answers are only logged. Training data and live data, side by side, with one input that does not match.](https://learn.benibix.com/diagrams/ai7-serving-path.svg)

---

Source: https://learn.benibix.com/term/ai/shadow-deployment
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-17
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
