The rapid rise of generative artificial intelligence (AI) poses profound challenges for libraries, archives, and museums (LAMs), institutions traditionally serving as stewards of knowledge and memory. Addressing these challenges was the focus of a recent webinar hosted by the Digital Library Federation (DLF), titled Content Authenticity and Provenance in the Age of Artificial Intelligence: A Call to Action for the LAMS Community. The session featured insights from Kate Murray of the Library of Congress and Joshua Sternfeld, authors of a report designed to help LAM professionals preserve public trust in the digital age.

Why Content Authenticity and Provenance Matter

As AI advances, enabling the seamless generation and alteration of digital content, the concepts of authenticity and provenance become increasingly critical. According to Murray, content authenticity refers to the verification that digital content is genuine and unaltered—though it does not ensure factual correctness. Provenance, on the other hand, involves documenting the origin and history of digital content, including any modifications it has undergone. Together, these principles form the foundation for maintaining public trust in the digital realm.

Murray emphasized that while these are not new concepts for the LAM community, the rapid advancements in AI demand a reevaluation of how authenticity and provenance are documented and communicated. In her presentation, she noted the growing public skepticism about digital content, particularly as instances of "deepfakes" and AI-generated misinformation continue to emerge. "Some media consumers are starting to think that all data and content are AI-generated unless told otherwise," Murray observed. This shift underscores the urgency for institutions to adapt their practices to ensure trustworthiness in the AI era.

The Report's Four Pillars for Action

The foundational report discussed during the webinar outlines four key pillars designed to guide the LAM community in addressing these emerging challenges:

1. Research: Institutions must prioritize research into new and evolving AI technologies and their implications for content authenticity. Understanding AI's potential benefits and risks is crucial for developing effective strategies and solutions.

2. Collaboration: The complexity of these issues necessitates cross-sector collaboration. Partnerships between LAMs, government bodies, technology developers, and other stakeholders can drive innovation while ensuring that institutional values and public trust remain central.

3. Advocacy: Proactively advocating for standards and policies that prioritize authenticity and provenance can help establish a framework for responsible AI use. This includes influencing the development of tools that align with the ethical and practical needs of LAMs.

4. Open Sharing: Transparent communication about methods, use cases, and challenges fosters collective learning. Sharing successes and failures helps the broader community adapt and refine its approach to emerging technologies.

Insights on Practical Tools and Technologies

Two specific technologies were highlighted in the webinar as examples of how LAMs can address issues of authenticity and provenance:

1. C2PA and Content Credentials: The Coalition for Content Provenance and Authenticity (C2PA) has developed content credentials, which provide cryptographically signed data revealing a digital object's origin and editing history. The Starling Lab—a collaboration between Stanford University and the USC Shoah Foundation—demonstrates one way this technology can be applied. By encrypting digital assets and their metadata at the point of capture, the Starling Lab ensures integrity through blockchain-based decentralized storage.

2. Model Context Protocol (MCP): Originally developed by Anthropic and overseen by the Linux Foundation, MCP acts as a gateway between large language models (LLMs) and trusted datasets. It connects LAM collections data to AI models like ChatGPT or similar systems, providing reliable provenance data and reducing the risk of distorted or hallucinated outputs. Sternfeld explained that this protocol represents "the first step of the dream scenario in which a researcher can use a computer to chat with a historical collection." Though promising, MCP is not without limitations and requires further development to minimize errors.

Challenges Ahead

The webinar also identified three key challenges that LAMs must address:

1. Institutional Capacity: The adoption of AI demands significant investments in staff training, hardware upgrades, and interoperable systems. Institutions must also address data privacy and security issues inherent in AI-assisted processes.

2. Ethical and Privacy Risks: Provenance data, particularly when AI is involved in content enhancement or reconstruction, can inadvertently reveal sensitive or proprietary information. Without careful human review, automated systems may generate misleading or unethical representations of cultural heritage materials.

3. Rapid Evolution of AI: The pace of AI development, coupled with the broader societal adoption of these technologies, creates significant uncertainty. As Sternfeld noted, "Practices need to keep up with constantly evolving technical landscapes while safeguarding transparency and trust."

Sustaining Public Trust

Despite these challenges, the report and accompanying conversation at the webinar underline the importance of deliberate and sustained action. Murray emphasized that the principles of authenticity and provenance must evolve without compromising the foundational values of the LAM community. "If LAMs lose our status as carriers of trusted information, we do not fulfill our mission, and no good comes from that," she cautioned.

The webinar discussed practical action items, including participating in community groups like those organized by the Library of Congress, which meet quarterly to share updates and use cases. Additionally, institutions were urged to engage with C2PA, MCP, and other emerging standards to learn how these tools can align with their preservation goals.

Moving Forward

The discussion concluded with an acknowledgment of the road ahead. Addressing authenticity and provenance in the age of AI is neither simple nor immediate. However, by fostering research, collaboration, advocacy, and transparent practice, the LAM community can rise to meet the challenge. As Sternfeld put it, "This moment demands deliberate, thoughtful, and sustained efforts to document and verify content throughout the digital preservation lifecycle."

The Digital Library Federation’s webinar serves as an important reminder that while AI poses unprecedented challenges, it also offers opportunities for innovation and resilience in preserving the integrity of cultural heritage. By acting now, libraries, archives, and museums can ensure that trust remains their most valuable asset.