Teachable Machine Privacy: How Your Images and Audio Are Handled

Google Teachable Machine does not upload your images or audio to Google’s servers. All training and testing happens locally in your web browser, so your data never leaves your device unless you explicitly export the model or share the project link. This article explains exactly how the tool handles your files, what happens when you save or share a project, and what privacy limits you should know about.

What Is Google Teachable Machine?

Teachable Machine is a free web-based tool from Google that lets you train a simple machine learning model using your own images, sounds, or poses. You create classes (for example, “cat� and “dog�), capture samples with your webcam or microphone, and click Train. The tool builds a neural network in your browser using TensorFlow.js. Once trained, the model can classify new inputs in real time. It is designed for beginners, educators, and hobbyists who want to experiment with machine learning without writing code.

How Teachable Machine Handles Your Data

According to Google’s official documentation, Teachable Machine processes everything client-side. That means the web app runs JavaScript in your browser, and the training data—your images, audio clips, or pose keypoints—are stored in your browser’s memory and used only for the local training process. The tool does not send your raw samples to Google’s servers for training. This is a key difference from many cloud-based ML services, where you upload data to a remote server.

However, there are important nuances:

• Webcam and microphone access: The browser will ask for permission to use your camera and microphone. That permission is granted to the page, and the data is processed locally. But the page itself could be modified by Google or a third party if the code changes, so it’s wise to check the current privacy policy.
• Model export: When you export your trained model, you download a file (e.g., a TensorFlow.js model) that contains the learned weights. That file is yours, and you control where it goes. If you upload it to a cloud service, that service will see it.
• Project sharing: Teachable Machine lets you save a project to Google Drive or download a project file. If you share a project link, anyone with the link can see the samples you used, because the project file contains the training data (images, audio, or pose keypoints). So sharing a project is effectively sharing your original data.

What Happens to Your Images and Audio After Training?

After you train a model, the raw samples are still in your browser’s memory until you close the tab or clear your browser data. They are not automatically uploaded anywhere. If you don’t export or share, the data is gone when you leave the page. The trained model itself is a set of numbers (weights) that represent patterns, not the original images. But if you export the model, the model file does not contain the original images—only the learned parameters. However, the project file (which you can download or save to Drive) contains the original samples, so treat that file with care.

Privacy Considerations for Different Use Cases

Privacy depends on how you use the tool:

• Personal experimentation: If you use your webcam and never export or share, your data stays on your computer. This is the most private scenario.
• Classroom or workshop: If you share a project link with students, everyone can see the training images and audio. If those contain faces or voices, that could raise privacy concerns, especially for minors.
• Commercial or sensitive data: If you train on proprietary or sensitive images (e.g., product designs or medical photos), remember that the project file contains those images. Storing it on Google Drive means Google’s servers will host it, subject to Google’s privacy policy.

Common Misconceptions

One common misconception is that Teachable Machine uses your data to improve Google’s models. That is not true for the local training process. Google does not see your samples unless you save the project to Google Drive or share the link. Another misconception is that the exported model contains your original images. It does not—only the learned weights. But the project file does contain the originals, so don’t confuse the two.

How to Protect Your Privacy

If you want to minimize privacy risks:

• Use images that don’t contain sensitive information. For example, use a stuffed animal instead of your pet’s face if you don’t want to share that.
• If you need to share a project, consider blurring faces or removing metadata from images before uploading.
• After training, clear your browser cache and site data to remove any residual samples from memory.
• Be cautious about saving projects to Google Drive if you are concerned about Google’s access. You can download the project file and store it locally instead.

Alternatives and Comparisons

Teachable Machine is not the only tool of its kind. Alternatives include:

• Lobe (by Microsoft): A desktop app that also trains locally, but it requires installation and is Windows/macOS only.
• Machine Learning for Kids: A web-based tool that uses IBM Watson, which does send data to the cloud.
• Custom TensorFlow.js: For developers, writing your own code gives you full control over data handling.

Compared to cloud-based services like Google Cloud AutoML, Teachable Machine is far more private because it doesn’t require uploading data. But it also has less power and flexibility.

Frequently Asked Questions

Does Teachable Machine store my images on Google’s servers? No, not during training. The images are processed locally in your browser. They are only uploaded if you save the project to Google Drive or share the project link.

Can Google see my webcam or microphone when I use Teachable Machine? The browser gives the page access to your camera and microphone, but the data is processed locally. Google does not receive the raw feed unless you explicitly save or share the project. However, the page code could change, so review Google’s privacy policy periodically.

What is the difference between the exported model and the project file? The exported model contains only the trained weights (the “brain� of the model). The project file contains the original training samples (images, audio, pose keypoints) and the model. Sharing the project file shares your original data.

Is Teachable Machine safe for use with children? It is safe in the sense that data stays local, but if you share a project link, the children’s faces or voices could be visible to anyone with the link. Always blur faces or use non-identifiable data in educational settings.

Can I use Teachable Machine offline? Yes, after the page loads, you can disconnect from the internet and it will still work, because training happens locally. However, you need an internet connection to load the page and to save to Google Drive.

What happens to my data if I close the browser tab? The training data is lost. Unless you exported or saved the project, there is no copy anywhere. That is a privacy feature, but it also means you can’t recover unsaved work.

Related

• Change Photos Backup Quality: Full Guide
• Grow a Garden Liberty Lily: Full Guide
• Fix iPhone Clock: Change Date, Time, or Time Zone