New AI transparency rules are making generated content easier to identify, but provenance presents a broader challenge: helping people establish where authentic digital material came from. This article explores what that means for cultural heritage organizations.
On August 2, 2026, new transparency requirements under Article 50 of the European Union’s AI Act began applying to certain AI systems and AI-generated content. Among other requirements, providers of generative AI systems must enable relevant AI-generated or manipulated content to be identified in a machine-readable format, while organizations using AI systems professionally have disclosure obligations for deepfakes and certain other AI-generated material.
The rules are an important step toward helping people understand when the content they encounter has been created or manipulated by AI. They also point to a bigger challenge.
As convincing synthetic images, audio, video, and text become easier to create, knowing that something is artificial is only one part of establishing trust. We also need reliable ways to understand where authentic digital content came from, who is responsible for it, and what has happened to it along the way.
For libraries, archives, historical societies, museums, and other custodians of primary sources, that makes provenance increasingly important.
The EU is one of the clearest examples of AI-content transparency becoming a legal requirement, but it is part of a wider international shift.
China introduced rules in 2025 requiring explicit and technical labeling for certain AI-generated text, images, audio, video, and virtual scenes. Canada is also examining greater transparency around AI-generated content and AI interactions as part of Canada’s National Artificial Intelligence Strategy: AI for All, launched in June 2026. In the United States, the approach remains more fragmented, although NIST is developing work around content provenance, authentication, watermarking, and synthetic-content detection, while proposed legislation such as the Protecting Consumers from Deceptive AI Act and the COPIED Act would establish stronger federal standards for identifying and tracing AI-generated content.
While approaches differ, the underlying issue is the same: as synthetic content becomes more common, people need better ways to understand the origins of what they encounter online.
An AI label can answer an important question: Was this generated or manipulated using AI?
Provenance addresses another: Where did this content come from?
This distinction matters for digitized historical material. A photograph might begin in a trusted institutional collection, supported by a catalog record, creator information, dates, descriptive metadata, and a clear connection to the organization responsible for the source material.
Once downloaded and shared elsewhere, however, that context can quickly disappear. The further an image travels from its source, the harder it can become to determine where it originated, whether it has been altered, and which version should be treated as authoritative.
That problem existed long before generative AI. Libraries, archives, and other cultural heritage institutions have long preserved provenance through cataloging, metadata, documentation, and responsible stewardship. AI simply increases the urgency by making convincing new and altered material much easier to produce.
The challenge is therefore not only to identify synthetic content. It is also to make the provenance of authentic content easier to establish.
Related reading: AI Is Making Trust the Most Valuable Asset in Digital Collections
Content Credentials offers one emerging approach. Based on the open C2PA standard, it is designed to create an interoperable provenance framework that can travel with digital media.
Content Credentials can attach digitally signed provenance information to digital media, including information about where content came from, when it was created, and aspects of its editing history. The underlying C2PA standard provides an open technical framework for recording and verifying the origins and changes associated with digital content.
For digital collections, however, the concept is compelling.
Today, much of the information that establishes the provenance of a digitized photograph, newspaper page, map, or other historical item sits in the metadata and descriptive information that surrounds it. When the digital object is downloaded or shared elsewhere, that connection to the institution and authoritative source can weaken or disappear.
Emerging provenance technologies could help preserve some of that connection as the digital item is shared beyond the collection where it originated.
We are looking beyond the labeling of AI-generated content and exploring how emerging provenance approaches such as Content Credentials could apply to digital collection workflows.
For us, the more interesting long-term direction is strengthening the connection between authentic digital objects, their provenance, and the institutions responsible for them as those objects move beyond their original source context.