Thought Leadership
Meta is enticing users with discounted prices for its AI model Muse Spark if, in return, they agree that their interactions with the AI may be used for training new models.
Users fundamentally have the right to object to the use of their own interactions for AI training. However, Meta is now specifically offering a discount to users who instead consciously opt to make their data available for training. In the regular pricing tier, using Muse Spark costs 1.25 US dollars per million input tokens and 4.25 US dollars per million output tokens. In contrast, users who agree to data usage pay only 10 US cents per million input tokens and 20 US cents per million output tokens in the so-called contributor tier.
For Meta, this model opens up an opportunity to obtain authentic original data that is valuable for its own AI training—a problem that other AI companies are also increasingly struggling with.
Meta: Previous employee tracking program already paused
The new initiative follows a previously discontinued, significantly more far-reaching project: Meta’s Model Capability Initiative program, introduced in April, temporarily recorded keystrokes, mouse movements, and screen content of its own employees on company-owned applications and websites, with the goal of training AI models to interact with computers in a way similar to humans. The program was suspended in June 2026 after an internal data leak potentially exposed private conversations, performance metrics, and meeting minutes of employees.
The current push thus joins a series of unusual approaches by various companies to obtain unique training data for their own AI models. Previously, for example, it was reported that Amazon was specifically buying up rare books, which were subsequently taken apart at a company-owned AI training facility in Las Vegas for the purpose of digitization, raising concerns among observers that the company might actually be destroying valuable historical books in the process.
(Editorial Team)
