RSL Collective—written evidence (AIC0023)
House of Lords Communications and Digital Select Committee inquiry: AI and copyright
The RSL 1.0 Standard: A Unified Rights and Licensing Framework for Implementing Copyright and AI Policy Across the Open Web
The RSL 1.0 Standard is an open web standard designed to enable the implementation of copyright and artificial intelligence policy objectives for lawful data use, rights control, transparency, auditability, and licensing across the public web.
While copyright law and future AI policy frameworks define what is required for lawful AI training and use, RSL provides a globally interoperable, open protocol built on the web’s existing architecture that defines how those requirements can be implemented at internet scale. RSL is designed to balance the needs of rights holders, AI developers, and regulators in line with the UK Government’s three stated objectives:
1. Control & Remuneration: Enabling rights holders to assert control over how their content may be used by AI systems, including the ability to license permitted uses and seek remuneration for AI training and other AI-related use cases.
2. Access & Licensing: Enable AI developers to access and automate licensing of large volumes of online content required to train their models easily, lawfully, and without infringing copyright
3. Transparency & Compliance: Provide transparency to rights holders and regulators about works used to train AI models, while enabling AI systems to demonstrate compliance by documenting permissions across billions of online content sources.
RSL does not require regulators to mandate a technical standard. It is an open, non-exclusive implementation layer that can support a range of legal outcomes determined by regulators, including different approaches to rights control, licensing, and remuneration. By separating policy decisions from technical implementation, RSL enables regulatory objectives to be implemented consistently and transparently across the open web without prescribing substantive legal rules.
From a recent article about RSL 1.0 by Shelly Palmer, CEO of The Palmer Group, a well-known media, technology, and digital economics expert:
“[RSL] provides a uniform way to declare rights in a world where AI systems rely on scaled access to text, images, audio, and code. It also creates a signaling layer for policymakers who are trying to balance copyright, fair use, innovation, and compensation. RSL puts publisher terms in a consistent format and creates a clear place for the industry to look."
"Courts are deciding whether training on publicly accessible material qualifies as fair use. Legislatures are examining what a licensing economy for AI training might require. RSL provides a structured vocabulary for those debates.”
To date, much of the standards work in this area has focused on opt-out controls that allow rights holders to prohibit certain AI uses. Control alone, however, is not sufficient to sustain a healthy content ecosystem in the AI era. Rights holders also need a practical way to offer licenses and receive remuneration, while AI developers require a reliable, automated mechanism to lawfully access and obtain licenses for large volumes of online content, rather than manually negotiating individual licenses with millions of disparate rights holders.
A similar limitation exists in widely used tools such as robots.txt. While robots.txt has long played an important role in managing crawler access, it was not designed to address modern AI use cases. In practice, it forces a binary choice: allow access for indexing and discovery, or block access entirely. For many publishers and other rights holders, reserving rights against AI training or AI-generated summaries now integrated into search results can mean losing visibility in traditional web search itself.
RSL builds on, rather than replaces, existing web infrastructure opt-out controls such as robots.txt by adding a complementary, machine-readable layer for rights expression and licensing. It allows content to remain discoverable in traditional web search, while reserving or licensing other AI uses, including training and AI-generated summaries, under defined terms. By separating discovery from extraction and reuse, RSL enables economically sustainable outcomes rather than relying on blanket opt-outs.
● Really Simple Licensing (RSL) is an evolution of the widely adopted RSS standard, which is used by more than 30 million websites to provide a machine-readable framework for licensing content to third-party clients and crawlers in exchange for traffic.
● The RSL standard is maintained by the non-profit RSL Collective and was developed through an industry-led standardization process involving leading publishers, platforms, and open standards organizations, including Yahoo, Ziff Davis, Fastly, O’Reilly Media, and Creative Commons.
● The RSL 1.0 specification is fully developed and published as an official industry standard, with comprehensive documentation available at https://rslstandard.org/rsl.
More than 1,500 publishers, brands, technology companies, and standards organizations have announced support for RSL. These include:
● global infrastructure providers, publishing platforms, and standards bodies that operate at internet scale, including Cloudflare, Akamai, Creative Commons, and the IAB Tech Lab; and
● publishers that collectively distribute and manage content across billions of web pages and serve large global audiences, including Reddit, Yahoo, People Inc., Ziff Davis, Internet Brands, Quora, Stack Overflow, The Associated Press, Vox Media, USA Today, CNET, BuzzFeed, Fast Company, HuffPost, Medium, MIT Press, Inc., Serious Eats, WebMD, WikiHow, Mashable, Quizlet, Ranker, Travel & Leisure, Boston Globe Media, Adweek, and The Guardian.
● The RSL 1.0 standard is built on widely deployed internet technologies and is compatible with existing web standards, including robots.txt, HTTP, and RSS.
● RSL’s decentralized design supports large-scale, distributed deployment across the public web, without reliance on a single central authority.
Building on established web practices such as robots.txt, the RSL 1.0 standard provides a more granular, machine-readable way for rights holders to express how their content may be used by AI systems. This additional granularity is important because different AI uses raise different legal, economic, and public-interest considerations. For example, RSL allows controls to be defined by:
● AI usage type (e.g., AI training, AI-generated summaries, indexing, web search), allowing rights holders to distinguish between discovery, analysis, and downstream reuse.
● End-user type (e.g., commercial, non-commercial, education, government, personal), helping balance remuneration for commercial use with continued access for researchers, educators, and non-profit organizations.
● Geographic restrictions, enabling rights holders to align permissions with jurisdiction-specific copyright rules and regulatory requirements.
A number of standards organizations and industry groups are actively developing approaches relevant to copyright and AI. These efforts address important aspects of the broader challenge, but each focuses on a specific part of the overall problem. To date, no single standard is intended to support the full lifecycle required for automated, legally compliant AI use of online content at internet scale, including rights discovery, licensing, authorization, and auditability.
● Authenticity and provenance standards such as C2PA, Content Credentials, and ISCC focus on verifying the origin and integrity of digital assets and detecting modification. These standards play an important role in trust, attribution, and content integrity, but they do not address how content may be used by AI systems or under what legal or licensing terms.
● Opt-out and preference signaling standards such as IETF AI Preferences and TDMRep enable rights holders to indicate whether certain AI-related uses are permitted or prohibited. These mechanisms support notice and control, but they do not define licensing terms, remuneration rules, or automated processes for obtaining authorization when use is permitted.
● Rights expression frameworks such as ODRL and RightsML are designed to model rights policies in a structured, machine-readable form. They provide a flexible way to express licensing requirements and obligations, but they are not designed to operate as end-to-end systems for large-scale automated licensing, authorization, compliance documentation, or enforcement across the open web.
Table: Comparison of RSL with Related AI Standards
Standard | Focus | Authenticity | Control | Remuneration | Automated Licensing | Audit Trail | Scalable Compliance |
|---|---|---|---|---|---|---|---|
RSL | Unified access, rights, and licensing standard | No | Yes | Yes | Yes | Yes | Yes |
C2PA TDM Assertions | Content authenticity and provenance | Yes | Images only | No | No | No | No |
Content Credentials | Consumer-facing provenance signals | Yes | No | No | No | No | No |
ISCC | Content fingerprint / identifier | Yes | No | No | No | No | No |
JPEG Trust (IPR Rights Expression) | Draft metadata framework for embedding IPR information in media files | Partial | Images only | No | No | No |
|
IETF AIPrefs | AI use-case preference signaling | No | Yes | No | No | No | No |
TDMRep | EU text/data mining opt-out indicator | No | Yes | Partial | No | No | No |
TDM AI Protocol | Early proposal for AI training permissions signaling | No | Yes | No | No | No |
|
AI.txt / Spawning AI | Proposed crawler instruction file for AI agents | No | Yes | No | No | No |
|
Open Rights Data Exchange | Proposal to structure rights information in linked-data format | No | Partial | No | No | No |
|
ODRL | Rights expression vocabulary | No | No | Yes | No | No | No |
RightsML | Rights expression vocabulary | No | No | Yes | No | No | No |
Authenticity: Whether the standard can verify that a content asset is genuine, track its origin, or detect whether it has been altered.
Control: Whether the standard allows a rights holder to declare which AI-related uses (such as training, AI-generated summaries, inference, or text and data mining) are permitted or restricted for a given content asset.
Remuneration: Whether the standard enables rights holders to define compensation or payment requirements as part of licensing content for AI use.
Automated Licensing: Whether the standard supports large-scale licensing through automated authorization and license acquisition, and can generate machine-verifiable proof that permission was obtained.
Audit Trail: Whether the standard provides mechanisms for AI systems to record permissions and licensing events so they can be demonstrated during audits or regulatory investigations.
Scalable Compliance: Whether the standard enables implementation of AI data governance requirements, including data sourcing, consent, rights management, and documentation, at internet scale across large numbers of websites and content assets or URLs.
RSL 1.0 has been published as a stable industry standard and is attracting interest from a wide range of stakeholders across the content, technology, and AI ecosystems.
As the Government continues its work on copyright and artificial intelligence, the next step for the RSL Technical Steering Committee is to invite feedback and discussion on whether, and how, RSL could be useful as a technical layer to support the UK’s evolving policy objectives.
In particular, the RSL Technical Steering Committee welcomes views from rights holders, AI companies, regulators, and other interested parties on the role that standardized, machine-readable approaches to rights control, licensing, and transparency might play in balancing the needs of creators and AI developers. This feedback will help inform whether RSL can be a practical and appropriate tool to support future copyright and AI frameworks.
January 2026
6