A New Frontier in AI: Publishers Take Control of Data Rights

The AI industry is moving toward a structured approach to data licensing as concerns over unpermitted training data usage mount. Following the high-profile copyright settlement involving Anthropic and the ongoing litigations against companies such as Midjourney, there is a growing realization of the need for a licensing mechanism to avoid a potential deluge of legal challenges. In response, a consortium of technologists and web publishers has introduced a licensing protocol designed to offer clarity and scale to AI data usage.

The Real Simple Licensing (RSL) protocol, a collaborative effort led by Eckart Walther, co-creator of the RSS standard, is designed to provide machine-readable licensing agreements that AI companies can use to train models on web content. This initiative aims to establish a technical and legal framework that facilitates large-scale data licensing. The RSL protocol enables publishers to define the licensing terms for their content, including whether AI firms must secure a custom license or use Creative Commons provisions. These terms are then embedded into the "robots.txt" file of participating websites, making it easier to determine the usage rights of specific data sets.

To complement the technical infrastructure, the RSL team has also created the RSL Collective—a licensing organization inspired by ASCAP and MPLC. This entity is designed to handle royalty negotiations and collections on behalf of rightsholders. The collective includes prominent web publishers such as Yahoo, Reddit, Medium, and WebMD, which are already part of or support the initiative. The RSL Collective aims to offer rightsholders a centralized platform for managing licensing terms with multiple AI developers simultaneously.

A notable feature of the RSL initiative is the inclusion of publishers who already have existing licensing agreements, such as Reddit, which earns an estimated $60 million annually from Google for data usage. While these publishers can still negotiate individual licensing deals, the RSL Collective offers a viable alternative for smaller publishers that lack the leverage to negotiate independently. The collective aims to ensure that all rightsholders—regardless of size—are fairly compensated for the use of their content.

The implementation of RSL is not without challenges. Unlike traditional media, where it is straightforward to track usage (e.g., a song being played), AI training data usage is more complex. The difficulty of determining when a specific document has been incorporated into a large language model (LLM) is compounded when payment is requested per inference rather than through a blanket fee. RSL's creators argue, however, that AI firms are capable of managing these complexities, as existing licensing agreements have required similar reporting. The goal is to create a system that is “good enough” to ensure compensation without requiring perfect tracking.

The RSL initiative now faces the critical test of adoption by AI companies. While firms such as ScaleAI and Mercor demonstrate that high-quality data is worth paying for, the broader AI landscape has traditionally relied on freely accessible data sources like the Common Crawl. Whether RSL can shift this paradigm remains uncertain. The initiative’s success will depend on whether AI leaders, including those who have publicly supported the need for such a system, commit to integrating RSL into their operations.