AUTHENTICATING     DIGITAL EVI­DENCE IN THE AGE OF GENERATIVE AI

Rethinking Criminal Procedural Safeguards

Imane Djeffal

Legal Researcher

Faculty of Law / Department of Criminal Law

Mohamed Khider University of Biskra, Algeria

2026

Abstract

The rapid advancement of generative artificial intelligence (AI) has introduced unprecedented challenges to the authentication of digital evidence in criminal proceedings. While traditional evidentiary frameworks have developed mechanisms to verify the origin, preservation, and integrity of digital materials, these mechanisms were primarily designed for an environment where digital content was assumed to originate from identifiable human actions or reliable recording devices. Generative AI has transformed this assumption by enabling the creation of highly realistic synthetic images, videos, audio recordings, and documents that may appear authentic despite not representing actual events. This development raises fundamental questions regarding the ability of criminal procedural systems to distinguish genuine evidence from artificially generated or manipulated content. This article examines whether existing criminal procedural safeguards remain sufficient to ensure the authenticity and reliability of digital evidence in the age of generative AI. Through a doctrinal and comparative legal analysis, the study explores the limitations of traditional authentication methods and evaluates emerging approaches to AI­related evidentiary challenges. The article argues that digital evidence authentication requires a transition from a purely integrity­based approach toward an AI­aware authenticity framework. It proposes the AI­Aware Digital Evidence Authentication Framework (ADEAF), based on five principles: provenance assessment, AI involvement disclosure, forensic verification, adversarial examination, and judicial reliability assessment. The article concludes that technological advancement should not weaken fundamental criminal procedural guarantees. Instead, legal systems must develop adaptive safeguards capable of preserving evidentiary reliability, protecting fair trial rights, and maintaining public confidence in criminal justice.

Keywords: Generative Artificial Intelligence; Digital Evidence; Evidence Authentication; Criminal Procedure; Digital Forensics; AI­Generated Content; Deepfakes; Fair Trial Rights; Artificial Intelligence Governance.

Table of Contents

1. Introduction ……………………………………………………………………………….. 1
2. Conceptualizing Digital Evidence Authenticity ……………………………………………… 2
   • Digital Evidence: From Electronic Records to Synthetic Content ……………………….. 2
   • Distinguishing Authenticity, Integrity, and Reliability ………………………………….. 2
   • The Need for an Expanded Concept of Authenticity ……………………………………. 3
3. Generative AI and the New Challenges of Digital Evidence Authentication …………………… 3
   • The Rise of AI­Generated and AI­Manipulated Evidence ……………………………….. 3 3.2. Deepfakes and the Challenge of Visual and Audio Authenticity ………………………… 3
   • The Limitations of Traditional Authentication Methods ………………………………… 3
   • Reconsidering the Legal Meaning of Digital Authenticity ………………………………. 4
4. Criminal Procedural Risks of AI­Generated Evidence ……………………………………….. 4
   • Threats to the Presumption of Innocence ………………………………………………. 4
   • The Burden of Proof and Beyond Reasonable Doubt ……………………………………. 4
   • The Risk of Wrongful Convictions …………………………………………………….. 4
   • Equality of Arms and the Right to Challenge Digital Evidence ………………………….. 4
   • The Need for a Human­Centred Approach to AI Evidence Assessment …………………. 4
5. Comparative Legal Perspectives ……………………………………………………………. 5
   • The United States: Flexible Authentication Standards and Judicial Evaluation …………… 5
   • The European Approach: Transparency, Accountability, and Risk­Based Regulation . . . . . . . . 5
   • Comparative Lessons and Policy Implications ………………………………………….. 5
6. Policy Recommendations: Building Safeguards for AI­Related Evidence ……………………… 5
7. Conclusion ………………………………………………………………………………… 6
8. References …………………………………………………………………………………. 6

1.   Introduction

Digital evidence has become an essential component of modern criminal justice systems. From electronic communications and mobile phone data to surveillance recordings and online activities, digital information increasingly influences criminal investigations, prosecutions, and judicial decisions [1]. The ability of digital evidence to reconstruct events and establish connections between individuals and criminal conduct has made it one of the most valuable forms of proof in contemporary criminal proceedings.

However, the rapid development of generative artificial intelligence has introduced a new challenge to the reliability of digital evidence. Unlike traditional forms of digital manipulation, generative AI systems are capable of producing synthetic content that can closely imitate reality. AI­generated videos, manipulated audio recordings, fabricated images, and artificially created documents can now reach levels of realism that make human detection increasingly difficult [2].

This technological transformation creates a fundamental legal problem: criminal courts have historically relied on the assumption that digital evidence represents a trace of an actual event, even when questions regarding its integrity or collection process exist. Generative AI challenges this assumption by creating circumstances where digital content may not merely be altered but may have never represented reality in the first place.

Traditional authentication procedures, including chain of custody, metadata examination, and expert verification, remain important safeguards [3]. Nevertheless, these mechanisms were developed primarily to determine whether digital evidence has been preserved without alteration after collection. They were not designed to address situations where the content itself may have been artificially generated before entering the criminal justice system.

The consequences of admitting unreliable AI­generated evidence are particularly serious in criminal proceedings. A false digital representation may influence investigative decisions, affect judicial evaluation, and contribute to wrongful convictions. Conversely, excessive skepticism toward digital evidence may prevent legitimate evidence from being effectively used in criminal prosecution [4].

Therefore, the challenge posed by generative AI is not simply technological but fundamentally procedural and legal. Criminal justice systems must reconsider how authenticity is understood and evaluated when digital evidence can no longer be assumed to represent a direct record of reality. This article argues that existing criminal procedural safeguards require adaptation to address the new evidentiary environment created by generative AI. It proposes an AI­aware approach to authentication that integrates technological verification with legal principles of fairness, transparency, and judicial protection.

The article addresses the following central question: Are existing criminal procedural safeguards sufficient to ensure the authenticity and reliability of digital evidence in the age of generative artificial intelligence? To answer this question, the article examines the evolution of digital evidence authentication, analyzes the procedural risks created by AI­generated content, compares emerging legal approaches, and proposes a framework for future criminal evidence assessment.

2.   Conceptualizing Digital Evidence Authenticity

2.1.   Digital Evidence: From Electronic Records to Synthetic Content

The concept of digital evidence has undergone significant transformation as technology has become increasingly integrated into social, economic, and criminal activities [5]. Traditionally, digital evidence referred to information stored or transmitted through electronic systems, including computer files, emails, mobile device data, digital photographs, and electronic communications. In criminal proceedings, such evidence has been used to establish facts, identify suspects, reconstruct criminal events, and support judicial decision­making.

The legal acceptance of digital evidence developed around the assumption that electronic records, although technically complex, represented traces of real­world activities. The primary concerns of courts and investigators focused on whether the evidence was collected lawfully, preserved adequately, and protected against unauthorized modification [6]. Accordingly, authentication procedures were designed to establish a connection between the digital material presented before the court and the original source from which it was obtained.

However, the emergence of generative artificial intelligence has transformed the nature of digital evidence. Modern AI systems are no longer limited to analyzing or organizing existing information; they can create entirely new digital content. Images, videos, voices, and documents can now be generated without any direct connection to an actual event or human action. This development introduces a new category of evidentiary material: synthetic digital evidence. Unlike conventional digital evidence, which generally represents a record of something that occurred, synthetic evidence may represent a constructed reality produced through computational processes.

The distinction is legally significant. Traditional evidence assessment asks: “Has this digital evidence been altered after its creation?” Generative AI requires courts to consider a more fundamental question: “Did this digital evidence correspond to a real event at any point?” This shift changes the nature of authentication. The challenge is no longer limited to detecting manipulation but extends to determining whether the digital content has a genuine relationship with reality. Consequently, criminal procedure must adapt to a new evidentiary environment where the existence of a digital file does not necessarily prove the existence of the event it appears to represent.

2.2.   Distinguishing Authenticity, Integrity, and Reliability

A significant challenge in the legal analysis of digital evidence is the tendency to use concepts such as authenticity, integrity, and reliability interchangeably. Although these concepts are closely related, they represent distinct legal inquiries and must be carefully separated, particularly in the context of AI­generated content.

• Authenticity: Concerns whether evidence is what it claims to be. In legal proceedings, authentication requires establishing a sufficient connection between the evidence presented and the source, event, or person it allegedly represents [7]. In the context of generative AI, authenticity becomes more complex because digital content may appear genuine while being entirely synthetic.
• Integrity: Refers to whether evidence has remained unchanged and protected from unauthorized modification throughout its lifecycle. Traditional digital forensic procedures often rely on metadata analysis, digital signatures, hash values, and chain of custody documentation [8]. However, integrity alone cannot guarantee authenticity. A perfectly preserved AI­generated video may maintain complete integrity from the moment it was created, yet still fail to represent a real event.
• Reliability: Concerns whether evidence is sufficiently trustworthy to support legal conclusions. It involves evaluating the quality of the evidence, the methods used to obtain it, and the circumstances surrounding its creation and interpretation [9]. AI­generated evidence creates additional reliability concerns because synthetic content may imitate reality convincingly and detection methods may produce inaccurate results.

2.3.   The Need for an Expanded Concept of Authenticity

The traditional understanding of authenticity was developed during an era where digital evidence primarily functioned as a record of human actions or physical events. Generative AI challenges this model by introducing evidence that may be digitally real but factually artificial. Accordingly, criminal procedure requires an expanded concept of authenticity that incorporates:

• The origin of digital content.
• The possibility of AI generation or manipulation.
• The reliability of verification methods.
• The transparency of technological processes.
• The ability of opposing parties to challenge the evidence.

3.   Generative AI and the New Challenges of Digital Evidence Authentication

3.1.   The Rise of AI­Generated and AI­Manipulated Evidence

The development of generative artificial intelligence has fundamentally transformed the nature of digital content and introduced new challenges for criminal evidence assessment. Unlike traditional digital manipulation techniques, which generally required specialized skills and significant technical resources, modern generative AI systems enable the rapid creation of highly realistic synthetic content accessible to a wide range of users [10]. AI­generated images, deepfake videos, synthetic audio recordings, and artificially created documents represent a new category of digital material that challenges traditional assumptions regarding evidentiary authenticity. These technologies can reproduce human appearance, facial expressions, voices, and writing styles with increasing accuracy, creating content that may appear indistinguishable from genuine evidence.

3.2.   Deepfakes and the Challenge of Visual and Audio Authenticity

Among the most significant examples of AI­generated evidence are deepfakes, which refer to synthetic media created through artificial intelligence techniques capable of altering or generating realistic representations of individuals [11]. Deepfakes present a unique challenge because criminal justice systems have historically placed considerable evidentiary value on audiovisual materials. Video recordings and audio evidence are often perceived as objective representations of events because they appear to capture reality directly. However, generative AI challenges this perception. A realistic video image or voice recording may no longer guarantee that the depicted person performed the alleged action or made the alleged statement [12].

3.3.   The Limitations of Traditional Authentication Methods

Existing approaches to digital evidence authentication generally rely on several mechanisms, including chain of custody, metadata analysis, forensic examination, and expert testimony. These mechanisms remain important components of evidentiary assessment; however, generative AI exposes their core limitations:

1. Chain of Custody: Demonstrates preservation from collection, but does not establish that content was genuine beforehand [13].
2. Metadata Analysis: May provide creation insights, but metadata can be removed, altered, or prove insufficient [14].
3. Forensic Tools: Continuously evolving, but not infallible and prone to error as AI models advance [15].
4. Expert Testimony: Raises questions regarding accessibility, methodological transparency, and adversarial challenge [16].

3.4.   Reconsidering the Legal Meaning of Digital Authenticity

The emergence of generative AI requires a broader understanding of authenticity within criminal procedure, pivoting on three core inquiries:

• Origin: Where did the digital content come from, and how was it created?
• Process: Was artificial intelligence involved in generating, modifying, or analyzing the content?
• Representation: Does the digital material accurately correspond to an actual event, person, or circumstance?

4.   Criminal Procedural Risks of AI­Generated Evidence

4.1.   Threats to the Presumption of Innocence

The presumption of innocence requires that every accused person be treated as innocent until proven guilty through a fair and lawful process [17]. The increasing use of AI­generated digital evidence creates new challenges for maintaining this principle, particularly when synthetic content possesses a high degree of persuasive power. Judges, investigators, and other decision­makers may unconsciously attribute greater credibility to audiovisual materials because of the common assumption that images and recordings provide direct access to reality.

4.2.   The Burden of Proof and Beyond Reasonable Doubt

The criminal standard of proof requires the prosecution to establish guilt beyond reasonable doubt. If AI­generated or AI­manipulated evidence becomes increasingly difficult to distinguish from genuine material, courts may face uncertainty regarding whether such evidence satisfies the required standard of proof. The principle of beyond reasonable doubt requires courts to consider uncertainty rather than ignore it [18].

4.3.   The Risk of Wrongful Convictions

A fabricated digital recording may create a false narrative that appears objectively verifiable. Unlike traditional forms of false testimony, synthetic evidence can simulate visual and auditory information, potentially making it more persuasive and difficult to challenge. Preventing wrongful convictions therefore requires a proactive legal response and strict authentication standards [19].

4.4.   Equality of Arms and the Right to Challenge Digital Evidence

The principle of equality of arms requires that both prosecution and defence have a reasonable opportunity to present their arguments and challenge opposing evidence [20]. AI­related digital evidence challenges this principle because the technical resources required to detect synthetic content may be disproportionately available to the state, creating a digital divide in criminal proceedings.

4.5.   The Need for a Human­Centred Approach to AI Evidence Assessment

Technology may assist courts in identifying patterns, analysing data, and evaluating evidence, but legal decisions concerning guilt and responsibility remain fundamentally human decisions involving rights, fairness, and moral judgment [21].

5.   Comparative Legal Perspectives

5.1.   The United States: Flexible Authentication Standards and Judicial Evaluation

The United States provides one of the most developed legal approaches to the authentication of digital evidence. Under Federal Rule of Evidence 901, parties must provide sufficient evidence to demonstrate that an item is what it claims to be [22]. While flexible, American jurisprudence must expand its analysis to incorporate AI involvement, provenance verification, and the reliability of detection technologies [23].

5.2.   The European Approach: Transparency, Accountability, and Risk­Based Regulation

The European Union has developed a broader regulatory approach toward artificial intelligence based on principles of transparency, accountability, and risk management [24]. In the context of digital evidence, transparency requires understanding whether AI was involved, what system was used, how reliable it is, and what limitations affect its outputs [25].

5.3.   Comparative Lessons and Policy Implications

• Authentication standards must evolve from a focus on digital preservation toward a broader assessment of digital authenticity.
• AI involvement in evidence creation or analysis should be disclosed to ensure procedural fairness.
• Technical expertise should support judicial decision­making rather than replace it.
• Legal systems must develop clear procedures allowing both prosecution and defence to challenge the reliability of AI­related evidence [26].

6.   Policy Recommendations: Building Safeguards for AI­Related Evidence

1. Establishing Specialized Authentication Guidelines: Create structured criteria for identifying AI­generated content, assessing provenance, and evaluating judicial reliability [27].
2. Enhancing Disclosure Obligations: Mandate transparency when AI contributes to the creation, modification, or analysis of digital material [28].
3. Strengthening Digital Forensic Capacity: Promote independent forensic expertise capable of handling AI­generated media detection and uncertainty evaluation [29].
4. Developing Judicial Training: Equip judges with foundational knowledge of generative AI capabilities, limitations, and forensic verification [30].
5. Ensuring Access to Defence Resources: Provide the defence with independent experts, legal assistance, and technical disclosure to maintain equality of arms [31].
6. International Cooperation: Foster cross­border harmonization on authentication standards, forensic methodologies, and fundamental rights protection [32].
7. Balancing Innovation and Protection: Avoid both technological rejection and uncritical technological acceptance [33].

7.   Conclusion

The emergence of generative artificial intelligence represents a significant transformation in the nature of digital evidence used in criminal proceedings. Existing authentication procedures remain important but are no longer sufficient when digital evidence may be artificially generated from the beginning. This article has argued that criminal procedure requires an AI­aware approach to digital evidence authentication. The proposed AI­Aware Digital Evidence Authentication Framework (ADEAF) provides a structured method based on provenance assessment, AI disclosure, forensic verification, adversarial examination, and judicial reliability assessment. Maintaining justice in the age of generative AI requires a careful balance between innovation and human judgment, between efficiency and fairness, and between the search for truth and the protection of individual rights.

8.   References

• Andrew Murray, Information Technology Law: The Law and Society (5th edn, Oxford University Press 2023).
• Robert Chesney and Danielle Keats Citron, ‘Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security’ (2019) 107 California Law Review
• Stephen Mason and Daniel Seng (eds), Electronic Evidence and Electronic Signatures (5th edn, University of London Press 2021).
• Orin S Kerr, ‘Digital Evidence and the New Criminal Procedure’ (2005) 105 Columbia Law Review
• Daniel J Solove and Paul M Schwartz, Information Privacy Law (7th edn, Wolters Kluwer 2021).
• International Organization for Standardization, ISO/IEC 27037:2012, Information technology — Security techniques — Guidelines for identification, collection, acquisition and preservation of digital evidence.
• Federal Rules of Evidence, Rule 901, ‘Authenticating or Identifying Evidence’.
• Association of Chief Police Officers, Good Practice Guide for Digital Evidence.
• Paul W Grimm, Maura Grossman and Gordon V Cormack, ‘Artificial Intelligence as Evidence’ (2019) 19 Northwestern Journal of Technology and Intellectual Property
• Yisroel Mirsky and Wenke Lee, ‘The Creation and Detection of Deepfakes: A Survey’ (2021) ACM Computing Surveys.
• Luisa Verdoliva, ‘Media Forensics and DeepFakes: An Overview’ (2020) 10 IEEE Journal of Selected Topics in Signal Processing
• Chesney and Citron (n 2).
• Mason and Seng (n 3).
• Murray (n 1).
• Verdoliva (n 11).
• Grimm, Grossman and Cormack (n 9).
• United Nations, International Covenant on Civil and Political Rights (1966), Article 14; European Convention on Human Rights, Article 6, Right to a Fair Trial.
• Orin S Kerr (n 4).
• Solove and Schwartz (n 5).
• European Convention on Human Rights, Article 6.
• UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021).
• Federal Rules of Evidence, Rule 901; Lorraine v Markel American Insurance Co 241 F.R.D. 534 (D Md 2007).
• Grimm, Grossman and Cormack (n 9).
• Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act).
• Council of Europe, Guidelines on Artificial Intelligence and Data Protection.
• OECD, Recommendation of the Council on Artificial Intelligence (2019).
• National Institute of Standards and Technology (NIST), Artificial Intelligence Risk Management Framework (AI RMF 1.0) (2023).
• Regulation (EU) 2024/1689 (Artificial Intelligence Act).
• Verdoliva (n 11).
• NIST (n 27).
• European Convention on Human Rights, Article 6.
• UNESCO (n 21).
• Murray (n 1).