IRAIET Research: Building a Verifiable Record for AI-Generated Digital Objects

Patent Application: System and Method for Providing Verifiable Provenance and Chain-of-Custody Management of Digital Evidence for Artificial Intelligence-Generated Digital Objects in Decentralized Environments

Filing date: 31.07.2026 Inventor: Yulia Leshchenko Applicant: Yulia Leshchenko

Artificial intelligence has made the creation of digital content almost instantaneous.

Establishing what happened to that content afterwards is a much harder problem.

An image, video, audio file, document, dataset, model output or other digital object can pass through multiple AI systems, editing tools and distribution platforms. Along the way, metadata can disappear, formats can change and the original record of how the object was created can become difficult to verify.

The first IRAIET patent application addresses this problem, which is not simply identifying AI-generated content The regulatory conversation is moving quickly toward transparency.

Under Article 50 of the EU AI Act, transparency obligations applying from August 2, 2026 require certain AI providers and deployers to support the identification of AI-generated or manipulated content. The European Commission's guidance and Code of Practice specifically address marking and detection mechanisms, including machine-readable solutions.

This is an important development. But identification and provenance are not the same thing.

A label can tell us that content was generated or manipulated by AI. It does not necessarily establish:

• which system generated it;

• under what execution conditions;

• when the operation occurred;

• whether the object was subsequently modified;

• who had custody of it;

• whether the record has been altered;

• or whether the current version can be reliably connected to the original record.

  
  

That distinction becomes particularly important when digital content is used in IP disputes, licensing, publishing, compliance, investigations or other contexts where evidence needs to be trusted.

What system is proposed?

The patent application describes a provenance management architecture designed to create an evidence package around an AI operation.

The system can obtain execution-context evidence, create a provenance package connecting that evidence with the digital object, cryptographically sign the package and register a corresponding commitment in a distributed registry.

It also introduces a chain-of-custody mechanism in which subsequent custody events are recorded as signed state transitions. This makes it possible to identify missing, reordered or unauthorized events in the object's history.

Another important element is provenance recovery.

If metadata has been removed through compression, format conversion or another permitted transformation, the system can use a resilient fingerprint or other locator to identify a possible provenance record and then validate that association against cryptographic and policy-based evidence.

The architecture is also designed around a practical problem with putting evidence directly on a public blockchain.

Sensitive information — including prompts, personal information, trade secrets, model parameters or restricted evidence — may not belong on a public ledger.

The proposed approach therefore separates the confidential evidence package from the public cryptographic commitment used to verify it. It can also support selective disclosure, allowing a verifier to confirm a particular fact without receiving the entire underlying record.

Why this matters for IP

For intellectual property, provenance is increasingly becoming part of the evidence problem.

Consider a digital asset that has passed through several AI systems and human editors before being published, licensed or incorporated into another work.

If a dispute later arises, the relevant question may not simply be “Was AI involved?”

It may be:

What was the original object? What changed? When did each transformation occur? Who authorized it? And can we demonstrate that history reliably?

That distinction matters for copyright, licensing, digital evidence and ownership disputes.

The technical architecture described in this patent application is intended to make that history more resistant to metadata loss and easier to verify. Its stated technical effects include resilient verification after transformations such as format conversion, compression and cropping, as well as cryptographic linkage between an AI execution environment, a digital object, an originator or authorized agent, time and subsequent custody events.

From AI transparency to evidence infrastructure

The broader significance of this research is the shift from disclosure to verification.

The regulatory environment is increasingly asking AI providers and deployers to make the origin of synthetic content identifiable. The technical standards ecosystem is moving in a similar direction: C2PA, for example, is designed to allow creators and custodians to make verifiable assertions about the creation and subsequent actions associated with digital content.

The next layer is determining how those records can remain useful when digital objects are transformed, redistributed or separated from their original metadata.

That is the research question behind IRAIET's first patent application.

It is not about predicting what AI will become.

It is about building better ways to establish what actually happened to a digital object.

And as AI-generated content becomes part of commercial transactions, intellectual property portfolios, publishing workflows and legal disputes, that distinction is likely to become increasingly important.