Bright Data, a data-collection company, facilitates AI model training by scraping data from the internet using its residential proxy network. This network, comprising over 400 million home IP addresses, is sourced from an SDK embedded in consumer apps, including those on smart TVs. The SDK, often hidden in privacy policies, allows Bright Data to route web-scraping traffic through users’ devices, raising concerns about privacy and potential misuse.

The Bright SDK uses a WebSocket connection to communicate with a server, sending device information and receiving instructions for scraping jobs. The SDK employs two inspection bypasses, one for the control plane and one for the data plane, making it difficult for network-based security tools to detect. The SDK also has country-specific bandwidth thresholds and a cross-platform identity linkage feature.