This forum, which was moderated by Paul Sanders, who is Chief Delivery Officer at Silverchair, was held on 27 August 2026 and explored the rapidly evolving landscape of licensing scholarly and authoritative content to artificial intelligence (AI) systems. The panelists: Todd Toler, Practice Lead for AI in Scholarly Communications at Ithaka S&R; Todd Ware, Senior Vice President, Publishing Chief Publishing Officer at American College of Physicians; and Jeremy Little, Vice President Artificial Intelligence at Silverchair focused on the emerging opportunities while protecting intellectual property, attribution, provenance, brand value, and relationships with libraries and other institutional subscribers.

A central theme was the shift away from viewing AI primarily as a model-training use case toward ongoing runtime access to content. Panelists highlighted retrieval-augmented generation (RAG), Model Context Protocol (MCP), APIs, gateways, and aggregators as emerging mechanisms through which AI systems can access authoritative content. Unlike one-time training agreements, runtime access may create new licensing opportunities, but its broader significance may be in sustaining existing institutional and professional relationships as usage moves into AI-mediated workflows. The economics remain unsettled, particularly at scale.

The panel also emphasized the need to establish clear guardrails and licensing terms as this market develops. Important considerations include defining the value of content, preserving attribution and citation, controlling how much content is displayed, preventing unwanted sublicensing, obtaining meaningful usage reporting, and mitigating risks associated with hallucinations and potential brand erosion. RAG-based access was viewed as particularly promising because it can support citations and direct users back to the version of record.  Distributed and agentic architectures complicate that path: an answer may draw on content retrieved across multiple publisher systems, with no single platform controlling whether a source is ultimately cited, displayed, or followed. The challenge is not simply to make authoritative content retrievable, but to carry provenance and measurable attribution with it from retrieval through user engagement.

From a technical perspective, publishers are relatively well positioned to participate because much scholarly content already exists in structured formats. However, AI delivery requires infrastructure that extends beyond the traditional publishing environments.  Publishers need consistent, machine-addressable endpoints spanning discovery, entitlement, provenance, and usage reporting.

The forum also identified the need for standards and measurement gaps. Existing publishing metrics do not adequately capture AI-mediated use, where content may be retrieved repeatedly without generating conventional downstream engagement or attribution. One framework discussed during the panel, SPUR, separates what was retrieved, what grounded the answer, what was cited, what was displayed, and what the user ultimately engaged with. Questions remain around common approaches to chunking, metadata, provenance, attribution, and reporting of AI usage.

Another important issue was the relationship between AI licensing and existing library subscriptions. Panelists discussed the possibility that new AI access models could create an additional payment layer for institutions and raised questions about how publisher rights, library agreements, authentication, and AI-tool costs should align. The market may ultimately rely on publisher-controlled access points, aggregators, or a combination of approaches.

The Q&A session further explored licensing, public-access content, version-of-record concerns, and reiterated the need to educate users and institutions about responsible use of AI-enabled content tools.

Key Takeaways

•     AI licensing is moving toward runtime access. Ongoing retrieval through RAG, MCP, APIs, and related technologies may become more significant than one-time training licenses.
•     An opportunity to shape the market. Having clear rights, licensing terms, technical requirements, and governance policies can help retain control as AI use expands.
•     RAG offers potential advantages. It can provide more traceable access to authoritative content, preserve citations, and direct users to the version of record.
•     New infrastructure will be required. Traditional web publishing systems may need to be supplemented with AI-specific delivery, authentication, metadata, and access mechanisms.
•     Measurement needs to evolve. Conventional download and usage metrics do not adequately capture AI-mediated retrieval and downstream use.
•     Standards remain immature. Attribution, provenance, chunking, metadata, and AI usage reporting lack broadly shared approaches.
•     Libraries and institutions will be central to the conversation. Stakeholders will need to determine how existing subscriptions interact with new AI access and payment models.
•     The market is developing quickly, but publishers still have an opportunity to act deliberately. The closing message was that organizations should begin defining their AI licensing strategy, rights framework, infrastructure needs, and governance approach now rather than reacting after the market has matured.

CHORUS would like to thank our sponsors for their continued support of CHORUS Events. The recording can be found on the event page.