Google purchases the source code of apps from the Play Store to train AI

Google trains AI on real Android app code

One of the most effective ways to use large language models (LLM) today is program code generation. In this context, Google has started collecting source code from Android apps and using it in its AI training projects.

How it works
1. Confidential initiative

According to 404 Media, the company launches a “confidential” program that is not publicly announced. Within it, Google acquires source code from Play‑Market developers for which they are paid.

2. Letter to developers

Developers receive an email in which Google describes the project as a “pilot project providing confidential materials” and promises additional revenue from their apps.

3. Terms of use

The letter states that the obtained code will be used to improve Google’s tools and products intended for developers. The link in the message leads to a page about “partnerships to improve our AI products,” but the topic of AI itself is not mentioned there.

Why this matters
- Model training

Human-created content often serves as material for training AI. Websites are a classic example, but app code is also valuable.

- Payment for intellectual property

Android app code is intellectual property and usually not published openly. Google pays for the right to use it in its models.

Thus, Google creates a new data stream for AI training while receiving compensation from app developers.