When Algorithms Become Witnesses: The Evidentiary Status of AI-Generated Intelligence in International Disputes

Author: Mauli Chaudhary
Student, CHRIST (Deemed to be University) Delhi NCR

———————————————————————————————————-

Abstract

The increasing use of artificial intelligence in producing, analysing, and transforming digital information is creating new evidentiary challenges for international adjudication. International disputes increasingly involve satellite imagery, photographs, videos, open-source intelligence, and other digitally derived material. Generative artificial intelligence complicates this landscape by enabling evidence to be altered, reconstructed, enhanced, or generated in forms that may appear indistinguishable from authentic material. This raises a question extending beyond conventional concerns about authenticity: when an evidentiary object is substantially produced through an algorithm, what should an international tribunal regard as its source, and how should its reliability and probative value be assessed? This article examines whether existing approaches to authenticity, provenance, reliability, and evidentiary weight are adequate for AI-generated intelligence. It distinguishes AI-assisted, AI-enhanced, and genuinely AI-generated material and analyses the implications of each for international adjudication. It argues for a provenance-and-process-based approach under which tribunals assess the underlying source, degree of algorithmic intervention, verifiability, reproducibility, and independent corroboration before determining evidentiary weight. The article ultimately contends that the principal challenge posed by synthetic evidence is not merely the possibility of fabricated evidence, but also the resulting erosion of confidence in genuine evidence.

Keywords: Artificial Intelligence, Digital Evidence, International Adjudication, Evidentiary Reliability, Synthetic Evidence.

💡 3 Quick Takeaways

AI-assisted, AI-enhanced, and AI-generated evidence present distinct evidentiary challenges and should not be treated as interchangeable categories.
Authenticity of a digital file does not necessarily establish the authenticity of the event it depicts. Tribunals may need to examine the underlying source and the process through which the content was produced.
A provenance-and-process-based approach, supported by independent verification and corroboration, can help international tribunals assess AI-generated intelligence without automatically admitting or excluding it.

I. Introduction

A photograph traditionally raises a familiar evidentiary question: whether the image is authentic and accurately represents the event it purports to depict. The emergence of generative artificial intelligence complicates this assumption. An image, video, audio recording, or document may now be created or substantially transformed by an algorithm without directly corresponding to an underlying real-world event. In international disputes, where factual controversies may concern military operations, territorial activity, attribution, or violations of international obligations, the consequences of synthetic material extend beyond conventional questions of technological reliability.

International adjudication is already familiar with technologically mediated evidence. Proceedings before international courts and tribunals have involved satellite imagery, photographs, videos, technical reports, and other digitally derived material. In Oil Platforms (Islamic Republic of Iran v. United States of America), satellite imagery and other material formed part of the evidentiary record. Satellite imagery and technical assessments have also featured in proceedings concerning territorial and environmental disputes between Costa Rica and Nicaragua.¹

The emergence of generative artificial intelligence introduces a qualitatively different problem. Traditional digital evidence generally begins with an event or object in the physical world, which is subsequently captured, stored, transmitted, analysed, or presented digitally. AI-generated material can disrupt this sequence. An algorithm may enhance an existing image, reconstruct missing information, generate synthetic content, or produce a seemingly realistic representation without a corresponding physical event. Consequently, the question before a tribunal may no longer be limited to whether an evidentiary file has been altered. It may instead be whether the apparent subject of the evidence ever existed in the form presented.

This distinction reveals a tension between authenticity, provenance, reliability, and probative value. A file may be genuine as a digital artefact while its substantive content is synthetic. Conversely, an AI-assisted analysis may be produced algorithmically while remaining grounded in authentic underlying material. Treating these categories as legally identical risks obscuring the characteristics that determine evidentiary reliability.

This article examines how international courts and tribunals should assess AI-generated intelligence and whether existing approaches to digital evidence are adequate for this purpose. It distinguishes AI-assisted, AI-enhanced, and genuinely AI-generated material before examining the implications of each category for evidentiary assessment. It argues that the appropriate response should not be a blanket rule of admission or exclusion. Instead, tribunals should examine the provenance of the underlying material, the degree of algorithmic intervention, the transparency and verifiability of the process, the possibility of independent reproduction, and the availability of corroborating evidence when determining evidentiary weight.

The central question is whether international adjudication must move beyond a conventional chain of custody towards a more comprehensive inquiry into the chain of production of evidence.

II. Conceptualising AI-Generated Intelligence

The expression “AI-generated intelligence” encompasses materially different forms of technological intervention. Treating them as a single category risks obscuring the legal questions each presents. A distinction should therefore be drawn between AI-assisted, AI-enhanced, and AI-generated material. This distinction is important because international courts and tribunals already encounter digitally derived evidence, for which questions of authorship, authenticity, reliability, and chain of custody have arisen.²

A. AI-Assisted Intelligence

The first category consists of evidence in which artificial intelligence assists with the analysis or organisation of pre-existing material without creating the substantive content relied upon by a party.

For example, an algorithm may analyse a large collection of satellite images to identify changes in terrain, detect objects, or locate patterns in video footage. Similarly, AI may assist investigators in sorting open-source material, translating communications, identifying potential matches, or organising large datasets.

In these circumstances, the underlying evidentiary material remains independently identifiable. The AI system functions primarily as an analytical tool between the original source and the tribunal.³ The principal evidentiary questions therefore concern the reliability of the underlying material and the methodology through which the AI-assisted analysis was conducted.

This category represents a comparatively limited departure from conventional digital evidence. The algorithm may influence the interpretation of evidence, but it does not necessarily become the source of the evidentiary content itself.

B. AI-Enhanced Intelligence

The second category occupies a more difficult position. AI-enhanced intelligence involves the modification, reconstruction, or improvement of existing material.

Examples include increasing an image’s resolution, removing noise from an audio recording, reconstructing portions of a damaged image, enhancing a video, or processing material to make information that was previously difficult to perceive more apparent.

The legal difficulty arises when enhancement moves beyond improving the presentation of existing information and begins to introduce information that was not directly observable in the original material. At that point, the distinction between recovering information and generating information becomes difficult to maintain.

The relevant question is therefore not simply whether AI was used, but what epistemic contribution the technology made to the evidence. An algorithm that makes an existing feature more visible presents a different evidentiary problem from one that predicts, reconstructs, or fills an evidentiary gap. This distinction matters because digital-evidence practice already recognises the importance of understanding how evidence was created and processed.⁴

C. AI-Generated Intelligence

The third category presents a distinctive evidentiary difficulty. AI-generated intelligence refers to material in which an algorithm produces substantive content presented as information about a person, object, event, or circumstance.

A synthetic photograph may depict an event that never occurred. A generated audio recording may reproduce a person’s voice without that person having made the statement. A synthetic video may represent conduct that did not take place.

Such material disrupts the conventional sequence through which evidence is produced. Rather than originating in a recorded real-world event, the material may be generated algorithmically and subsequently presented as a representation of that event.

The distinction is fundamental. Where conventional digital evidence generally has an identifiable underlying event or source, genuinely synthetic material may not. A tribunal must therefore confront a preliminary question that differs from ordinary authentication: what, if anything, is the evidentiary relationship between the generated representation and the real-world event it purports to depict?

This does not mean that AI-generated material should automatically be excluded or that the use of AI necessarily makes material unreliable. Rather, the degree of algorithmic intervention becomes relevant to determining what can properly be inferred from the material and what evidentiary weight it should receive.

D. Why the Distinction Matters

These categories have different evidentiary implications. AI-assisted intelligence primarily concerns the analysis of existing evidence; AI-enhanced intelligence may alter or reconstruct that evidence; and AI-generated intelligence presents a further difficulty because the substantive content itself may be algorithmically produced.

The greater the role of AI in producing the evidentiary material, the greater the need to examine its source, method of production, reliability, and independent verifiability. This distinction provides the basis for assessing whether existing principles governing digital evidence are adequate to address AI-generated intelligence.

III. The Existing International Evidentiary Framework

Before assessing AI-generated intelligence, it is necessary to examine how international courts and tribunals presently evaluate digital evidence. Contrary to the assumption that international adjudication lacks evidentiary standards, international forums have long exercised discretion in admitting and weighing documentary, photographic, audiovisual, satellite, and expert evidence.⁵

The legal question is therefore not whether digital evidence can be admitted, but whether existing principles are sufficient when its content may be generated or materially reconstructed by artificial intelligence.

A. Authenticity and Provenance

International tribunals may distinguish questions concerning the admission of evidence from the assessment of its evidentiary weight. They generally examine whether material is authentic, whether its origin can be established, and whether its chain of custody or provenance supports its reliability.

Authenticity concerns whether evidence is what it purports to be, while provenance concerns its origin and subsequent handling. These principles are particularly significant in disputes involving digital photographs, satellite imagery, videos, and electronic records, where the source and integrity of the material directly affect its probative value.⁶

However, the use of AI introduces an additional difficulty. Even if a digital file is authentic and its custody is properly documented, those factors may not establish that its substantive content accurately represents the event in question.

B. Evidentiary Discretion of International Tribunals

International courts and tribunals generally exercise considerable discretion in admitting and assessing evidence, although the precise evidentiary framework varies between institutions.

The International Court of Justice has powers concerning the receipt of evidence and expert material under its Statute.⁷ International criminal tribunals similarly assess evidence under their applicable procedural frameworks, taking account of relevance, reliability, and probative value. Article 69 of the Rome Statute and Rule 63 of the International Criminal Court’s Rules of Procedure and Evidence are relevant to the treatment and assessment of evidence before the ICC.⁸

Technologically derived evidence is not necessarily excluded merely because it originates from digital or scientific methods. Its credibility is assessed within the applicable procedural framework and the factual context of the dispute.

This flexibility provides a foundation for addressing AI-generated intelligence. Nevertheless, flexibility alone may not resolve the difficulties that arise when the content itself has been generated or materially reconstructed by an algorithm.

C. Digital Evidence in Contemporary International Disputes

International proceedings increasingly involve digitally derived material, including satellite imagery, photographs, videos, and other forms of open-source information. These developments demonstrate that such material has become an important component of international fact-finding.

Its evidentiary value, however, continues to depend on questions of authenticity, provenance, and reliability.⁹ The growing use of digital evidence makes it increasingly important to distinguish the technological form of a piece of evidence from the process through which its substantive content was created.

D. The Emerging Evidentiary Challenge

Existing principles provide a basis for addressing many forms of digital evidence, but they were largely developed for material originating from an identifiable source. Generative artificial intelligence complicates this premise by producing or materially reconstructing content that may appear authentic despite lacking a conventional evidentiary origin.

The central issue is therefore not the admissibility of digital evidence itself. It is whether traditional concepts of authenticity, provenance, and probative value remain adequate when the evidentiary content is substantially shaped by algorithmic generation.

IV. The AI Evidentiary Problem: When Authenticity Is No Longer Enough

The existing framework becomes more difficult to apply when artificial intelligence participates in the creation or reconstruction of evidentiary material. The difficulty does not arise simply because AI is technologically complex. It arises because the conventional relationship between an event and the evidence representing that event may be disrupted.

A. The Problem of Authenticity

Traditional digital evidence generally allows a tribunal to ask whether the material presented is genuine and whether it accurately represents its purported source.¹⁰ AI-generated material complicates this inquiry because a digital file may itself be genuine while the content it depicts is synthetic.

Establishing the authenticity of the file does not necessarily establish the authenticity of the event represented in it. A file may have an identifiable creator, a documented storage history, and an intact chain of custody, yet contain a synthetic representation that does not correspond to an actual event.

The relevant question consequently shifts from “Is this digital file authentic?” to “Does the content of this file reliably correspond to the event or circumstance it purports to represent?”

B. The Problem of Provenance

Provenance becomes particularly significant where AI has modified or generated evidentiary material. A tribunal may need to determine the origin of the underlying material, the stages through which it was processed, and the nature of the AI system used to produce the final version.¹¹

This raises a further question: should the evidentiary chain extend beyond the custody of the material to include the process through which its content was generated or transformed?

For AI-generated intelligence, answering this question may require disclosure of the underlying source, the extent of algorithmic modification, and the methodology used to produce the material. A conventional chain of custody may establish who handled a file and when, but it may not adequately explain how the content itself came into existence.

C. The Problem of Reliability

AI-generated material may appear highly convincing while remaining factually inaccurate. The difficulty is compounded by the limitations of existing detection methods.

The United States National Institute of Standards and Technology (NIST) has observed that the performance of deepfake-detection systems can vary substantially depending on the data, generator, and processing conditions. Detection performance may become particularly challenging when systems encounter unfamiliar generators or content that has undergone post-processing.¹²

This creates an evidentiary dilemma for international tribunals. Reliance on technological detection cannot, by itself, resolve questions of authenticity where the reliability of the detection method must also be assessed.

The problem is therefore not simply that AI can produce false evidence. Synthetic media may also make genuine material more difficult to authenticate, shifting the inquiry from detection alone towards provenance, verification, and corroboration.

D. The Problem of Probative Value

These difficulties suggest that the use of AI should not automatically determine whether evidence is admitted or rejected. Instead, the extent to which AI contributed to the evidentiary material should inform the tribunal’s assessment of its probative value.

The crucial inquiry is not whether AI was involved, but what it contributed, whether that contribution can be independently assessed, and whether the resulting material can be corroborated by other evidence.

This distinction leads to the central proposition of this article: AI-generated intelligence may require international tribunals to look beyond the conventional chain of custody and examine the process through which the evidentiary content itself came into existence.

V. Towards a Legal Framework for AI-Generated Intelligence

The existing principles of authenticity, provenance, reliability, and probative value provide a starting point for evaluating AI-generated intelligence.¹³ Their application may, however, require greater procedural scrutiny where artificial intelligence has materially contributed to the evidentiary content.

Rather than creating a blanket rule excluding AI-generated material, international tribunals could assess such evidence through a structured inquiry.

A. Identifying the Underlying Source

The first inquiry should concern the source of the information. Where AI has processed an existing photograph, video, satellite image, or document, the original material should, where possible, remain available for examination.¹⁴

Where no identifiable underlying source exists, the tribunal may face greater difficulty determining what factual proposition the generated material can reliably establish. The absence of an identifiable source does not necessarily settle the question of admissibility, but it is relevant to assessing the relationship between the material and the fact it is offered to prove.

B. Examining the AI Process

The tribunal should next consider the role played by the AI system in producing the evidence. Mere analytical assistance presents a different concern from substantive reconstruction or generation.

Relevant considerations may include the nature of the system used, the extent of its intervention, and whether the process can be adequately explained or independently examined.¹⁵

Where a party relies materially on AI-generated or AI-enhanced evidence, procedural fairness may also require disclosure of sufficient information to permit a meaningful challenge. Depending on the circumstances, this may include the original material, the nature of the algorithmic intervention, and relevant information about the processing method.

Such disclosure would allow the opposing party and the tribunal to assess whether the resulting material can be independently verified. The precise scope of disclosure would depend on the applicable procedural rules and the circumstances of the case.

C. Independent Verification and Corroboration

AI-generated intelligence should, where practicable, be tested against independent sources of evidence. These may include contemporaneous records, witness testimony, satellite imagery, metadata, or other independently obtained material.¹⁶

Independent corroboration becomes particularly important where the original source cannot itself be examined or where the AI process involves reconstruction or generation.

Corroboration does not automatically establish that a disputed item is reliable. Rather, it allows the tribunal to evaluate whether the factual proposition supported by the AI-generated material is independently supported by other evidence.

D. Determining Evidentiary Weight

Ultimately, the presence of AI should not automatically determine the admission or rejection of evidence. Instead, the tribunal should consider the evidence’s provenance, production process, reliability, verifiability, and corroboration when determining its probative weight.¹⁷

This approach recognises that AI may both strengthen and undermine evidentiary reliability. It avoids treating technological involvement as either inherently trustworthy or inherently suspect.

The proposed framework would shift the inquiry from merely asking whether evidence has been properly preserved to also asking how its content was produced. This distinction is particularly significant where AI-generated material can appear authentic without necessarily corresponding to an underlying real-world event.

A possible objection is that AI-generated intelligence does not require a distinct evidentiary framework because international tribunals already assess evidence according to relevance, reliability, authenticity, and probative value. On this view, AI involvement merely introduces another factor for consideration rather than creating a new legal category.

This objection has force where AI is used only to analyse existing material. It is less persuasive where an algorithm materially generates the content presented as evidence. In such cases, assessing the evidentiary object may require scrutiny not only of its custody and authenticity, but also of the process through which its substantive content was produced.

The proposed approach need not displace existing evidentiary rules. It can instead operate as a structured method for applying established principles to a category of evidence that raises distinct questions about origin and production.

VI. When AI-Generated Evidence Conflicts with Human Evidence

The evidentiary difficulty becomes more complex when AI-generated intelligence conflicts with evidence derived from human sources.

A tribunal may, for instance, be presented with an AI-generated video that appears to depict an event while eyewitness testimony asserts that the event did not occur. The reverse situation is equally possible, where human testimony conflicts with technologically generated material.

Neither human testimony nor algorithmically generated material should be treated as inherently reliable merely because of its source. Human testimony may be affected by memory, perception, or deliberate falsehood, while algorithmically generated material may contain errors that are not apparent from its presentation.

The appropriate inquiry should therefore focus on the provenance, methodology, consistency, and corroboration of each evidentiary source.¹⁸ Where competing evidence cannot be conclusively reconciled, the tribunal should assess which account is better supported by independently verifiable material.

Such material may include contemporaneous records, metadata, satellite imagery, additional witness testimony, or other evidence capable of corroborating the disputed account.

The emergence of synthetic evidence therefore creates a distinctive evidentiary dilemma. AI does not merely provide a new means of producing potentially false evidence; it may also provide a basis for challenging genuine evidence as supposedly fabricated.

Consequently, the central challenge for international adjudication is to preserve the ability to distinguish credible evidence from merely convincing evidence, without treating either technological or human origin as conclusive of reliability.

VII. Conclusion

Artificial intelligence is altering not only the production of information but also the conditions under which information can be treated as evidence. International courts and tribunals already possess established principles concerning authenticity, provenance, reliability, and probative value. The emergence of AI-generated intelligence, however, may expose limitations in applying these principles where evidentiary content is algorithmically created or reconstructed.

The appropriate response is not to treat all AI-generated material as inherently unreliable or to create an automatic rule of exclusion. Instead, tribunals should examine the underlying source, extent of algorithmic intervention, production process, independent verifiability, and corroboration before determining the evidentiary weight of such material.

The central concern is not simply whether an algorithm can produce convincing evidence, but whether the legal process can determine what that evidence actually establishes. As synthetic material becomes increasingly difficult to distinguish from genuine recordings, international adjudication may need to supplement traditional inquiries into the custody and authenticity of evidence with greater scrutiny of its production and provenance.

Ultimately, the challenge posed by AI-generated intelligence is epistemic as much as procedural. When convincing evidence can be manufactured, preserving the ability to distinguish what is merely plausible from what is legally reliable becomes essential to international adjudication.

Endnotes

Oil Platforms (Islamic Republic of Iran v. United States of America), Judgment, 2003 I.C.J. Rep. 161, 188–89 (Nov. 6, 2003); Certain Activities Carried Out by Nicaragua in the Border Area (Costa Rica v. Nicaragua), Compensation, Judgment, 2018 I.C.J. Rep. 15, 40 (Feb. 2, 2018).
Daniel Brantes Ferreira & Elizaveta A. Gromova, Digital Evidence in Disputes Involving States, 118 AJIL Unbound 51, 51–56 (2024); Sofia Aalto-Setälä et al., Leiden Guidelines on the Use of Digitally Derived Evidence in International Criminal Courts and Tribunals (Kalshoven-Gieskes Forum on International Humanitarian Law 2022).
Aalto-Setälä et al., supra note 2.
Ferreira & Gromova, supra note 2, at 55–56; Aalto-Setälä et al., supra note 2.
Ferreira & Gromova, supra note 2, at 52–55; Statute of the International Court of Justice arts. 48–52, June 26, 1945, 33 U.N.T.S. 993.
Ferreira & Gromova, supra note 2, at 54–56; Aalto-Setälä et al., supra note 2.
Statute of the International Court of Justice arts. 48–50, June 26, 1945, 33 U.N.T.S. 993; Ferreira & Gromova, supra note 2, at 52–55.
Rome Statute of the International Criminal Court art. 69, July 17, 1998, 2187 U.N.T.S. 90; Rules of Procedure and Evidence, Rule 63, International Criminal Court.
Ferreira & Gromova, supra note 2, at 52–56; Aalto-Setälä et al., supra note 2.
Aalto-Setälä et al., supra note 2; Ferreira & Gromova, supra note 2, at 54–56.
Ferreira & Gromova, supra note 2, at 55–56.
Bilva Chandra et al., Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency, NIST AI 100-4, at 27–29 (National Institute of Standards and Technology 2024).
Ferreira & Gromova, supra note 2, at 54–56; Aalto-Setälä et al., supra note 2.
Aalto-Setälä et al., supra note 2.
Ferreira & Gromova, supra note 2, at 55–56.
Aalto-Setälä et al., supra note 2.
Ferreira & Gromova, supra note 2, at 53–56; Rules of Procedure and Evidence, Rule 63, International Criminal Court.
Rules of Procedure and Evidence, Rule 63, International Criminal Court; Ferreira & Gromova, supra note 2, at 54–56.

Disclaimer: The views expressed in this article are those of the author and do not necessarily reflect the views of The Lawscape.

The Lawscape — clear, practical legal insight for students and future lawyers.