Amazon Used Twitch Content for AI Training for Years—Now Streamers Can Opt Out

Amazon has been using content from its streaming platform Twitch to train its AI models for years, a practice that only recently became an opt-out for creators. According to Ars Technica, which first detailed the long-standing policy, the quiet data harvesting has been a feature, not a bug, of the Amazon-owned platform. The recent addition of an opt-out option comes only after significant public criticism from streamers who discovered their work was being used by default.

The core of the issue for many creators was the lack of transparency. Both BBC Business and BBC Technology reported on the backlash from users who felt their content was being repurposed without clear, affirmative consent. For a platform built entirely on the creative labor of its users, treating their content as a free R&D resource for the parent company has predictably eroded trust. This is the fundamental tension of user-generated content platforms: the users are both the customer and the product.

A Belated Concession

In response to the criticism, Twitch has updated its policies. A new toggle allows creators to prevent their user-generated content from being used for what the platform calls “future Gen AI model improvements.” The language is specific. It notably addresses future training, not past usage, leaving creators to wonder how many of their streams have already been fed into Amazon's models.

Ars Technica reports this opt-out is a new development, suggesting the company was comfortable with the default opt-in until public pressure mounted. This pattern is familiar in the tech industry. Platforms often operate under ambiguous terms of service, maximizing their access to data until users or regulators force a clarification. Offering an opt-out after the fact is a standard crisis management playbook, designed to quell dissent without fundamentally altering the company's strategic access to valuable data.

The Unpaid Data Workforce

The incident highlights a structural conflict in the race to build generative AI. These models require massive, diverse datasets to function, and the internet's largest platforms are sitting on goldmines of human expression. For Amazon, Twitch's millions of hours of live and archived video represent a uniquely valuable training set for developing more sophisticated AI. For streamers, however, it represents their intellectual property and their livelihood.

Together, these reports point to a system where creators are treated as an unpaid data workforce. While the new opt-out provides a measure of control, it places the burden on individual creators to discover and disable a setting that many believe should be opt-in from the start. The pattern indicates that for large tech companies, asking for forgiveness is still seen as a more efficient business strategy than asking for permission.