The blog has been written by Saumil Ahuja, a 4th-year student at Christ (Deemed to be University), Bangalore

The Rise of Deepfakes: A Threat Beyond Technology

In recent news, Preity Zinta went to the Bombay High Court seeking an injunction for breach of personality rights and alleging that the defendants had created, uploaded and made AI-generated videos defaming her. The visual effects used by the specialist created the footage so well that it was sufficient to evade the leading detection tools. If such content can make the casual eye fail, then certainly it has the potential to evade judicial detection and pose a greater challenge to the court’s evidentiary reliability, relying on traditional mechanisms to trace electronic records.

Deepfakes, being infamous for the notorious use of AI, are digital images, videos, or voices, etc. that have been so convincingly generated or altered so as to present someone doing an act which was not actually done or said by them. This term finds its origin in the synthesis of two words: deep learning and fake. Initially regarded as just a technological advancement in computer vision or neural networks, it has shown unprecedented progress with increased computing speed due to the widespread availability of  generative adversarial networks.  

The threat posed by deepfakes is no longer hypothetical, and their legal consequences are also not speculative. Recently, in the case of Mendones v. Cushman & Wakefield, a deepfake video was submitted as evidence and caught by California Judge Victoria Kolakowski. The challenge propounded against the court is not merely technological but jurisprudential as well, as the law of evidence framework is designed for something that has happened, and this presumption is now under siege. The real question is not whether deepfakes exist or not but rather how Indian courts respond.

The Evidentiary Crisis and Collapse of Traditional Assumptions

In the Indian jurisdiction, video and electronic evidence, though imperfect, have been considered reliable evidence by the courts. Deepfakes have shattered this presumption entirely. As Judge Herbert B. Dixon Jr. of the Superior Court of the District of Columbia remarked, “deepfakes are designed to gaslight the observer and any truism which has been associated with the archaic presumption of seeing is believing might disappear from our ethos.”

The challenge posed by it in an evidentiary crisis is mostly twofold. The foremost being that the party may submit a deepfake video as genuine evidence, for example, deepfakes containing incriminating statements, which the person in it never made. Secondly, the more subtle yet concerning issue is that the party may challenge even the authentic evidence by alleging that the evidence is AI- generated. The latter process is what scholars call the ‘Liar’s dividend’, as it allows unethical litigants to intentionally dispute the authentic evidence as fabricated, to inject doubt in the eyes of the court.

Adding to the challenges , the rise of agentic AI being used to summarise contracts or generate pleadings, considering the malleability of the content, means the existing rules of evidence were not designed for this. The identification is certainly a herculean task, as the visual irregularities, such as blurring or lagging, might be a result of low internet connectivity rather than deliberate AI manipulation. As quoted correctly by Prof. Daniel Linna[i], there is no foolproof way to identify the evidence as authentic or AI-fabricated, but these are not future challenges anymore; they have already knocked on the doors of courts.

Existing Indian Laws: No dedicated Legislation

India lacks comprehensive legislation dealing with the issue, though scattered legislations only offer indirect protection. Under the Indian Constitution, Articles 15, 19 & 21 offer indirect protection by prohibiting deepfake-related discrimination, misinformation and privacy-related violations. Other legislations like POCSO Section 13 penalises deepfake related sexual abuse content, and Sections 43 & 66D of the IT Act, 2000 indirectly address unauthorised data manipulation and impersonation. Yet, none of the existing legislation is specifically dedicated to the cause of deepfakes and their evidentiary value.

Global Response and Lessons from the USA and Canadian Legal Framework

The study of the USA and Canadian approaches is significant for analysing an adequate response towards deepfakes:

The USA currently operates under theFederal Rule of Evidence.Rule 901 propels the proponent to showcase that ‘it is more likely than not’ the evidence he claimed it to be. This threshold is being challenged as too low; therefore, Scholar John LaMonaga has proposed an amendment to the rule, as it would require the court to go beyond and not merely rely on the witness statements and allow the accused to request a hearing, propounding the proponent to furnish the corroborating evidence. In the case of Dadras International v. Iran, though long before the digital evidence was introduced, it emphasised a clear and convincing standard of proof in order to allege the evidence presented, thus considering a heightened standard of proof. The heightened standard adopted can pose dual possibilities, as it deters frivolous challenges but may embolden unethical actors, as the fabrication of evidence is difficult to prove.

In this development, California has taken the longest stride, as the legislation enacted through SB 970 mandates the Judicial Council to review the impact of fabricated evidence and to develop rules for evaluating authenticity concerns. Considering this, the courts have imposed sanctions on parties for the submission of AI-fabricated evidence, establishing that they will not tolerate AI-manipulated evidence. The proposed No Fakes Act aims to regulate the misuse of deepfakes.

In the Canadian jurisdiction, Article 31.1 places the burden squarely on the person propounding the digital evidence to prove the authenticity of the evidence that it is that which is purported to be, and there is no concept of shifting the burden of proof in it.

The Indian evidentiary vacuum: Towards a proportionate legal framework

The Indian evidentiary domain, now majorly governed by the provisions of Bharatiya Sakshya Adhiniyam, has taken significant steps in acknowledging electronic evidence, yet there is no dedicated provision for dealing specifically with AI-Generated deepfake content. Additionally, there is no provision for addressing authentication, no provision for proactively engaging in the evidentiary hearing on deepfake-related content and most importantly, no mechanism for providing the court with the methods to evaluate the concerns relating to metadata or chain of custody events specific to synthetic data. Therefore, Indian courts have not taken any proactive steps like the US and Canadian frameworks and thus, considering the delicacy, this silence is no longer a mere theoretical problem but a pending crisis.

The proposed framework for the Indian legislature is to strive towards proportionality and not paralysis. Considering the significant advantages of deepfakes in the field of education, healthcare, cinema, etc., regulating the misuse of deepfakes is a more viable option than a complete ban. Thus, Indian courts cannot expect the elimination of AI-fabricated evidence. Keeping in view the nature of the challenge posed, a proportionate authentication framework has to be developed.

Firstly, the party submitting the evidence has to bear the initial burden of proving the authenticity of the evidence and thus aligning with the Canadian model. Secondly, the opposing party alleging the fabrication of evidence cannot simply state that it is AI-generated; instead, the challenger must raise a prima facie case, alleging the specific anomalies, visible artefacts, or inconsistencies to oblige the court to act further on it. Anything less than that would open the floodgates, as it would be used as a delaying tactic for the proceedings. Thirdly, once a satisfactory challenge has been posed to the propounded evidence, enhanced authentication should be required from the proponent, which can include examination of metadata, provenance records, chain of custody and other crucial corroborative evidence. The courts should be in a position to pose questions related to authenticity, if any, without immediately ordering a forensic investigation. Lastly, the forensic investigation has to be reserved for high-stakes matters in which visible inconsistencies exist in artefacts, anomalous data, or any suspicious conduct is present.

The court must not order a forensic investigation as a matter of course in every case suffering an allegation involving deepfakes, as it would be highly impractical and time-consuming. The concern is not merely incidental but central in nature, as Hon’ble Justice Surya Kant has also emphasised the ardent need for timely justice, as he remarks that “Justice delayed is not justice denied but justice destroyed. Therefore, the evidentiary framework which turns every contested evidence into a forensic investigation will only compound the pendency rather than reduce it. Thus, the proportionality principle, constituting a healthy balance between adjudication accuracy and timely justice, will foster justice and eventually uphold the Rule of Law. 

Operationalising the framework: Decoding the applicability to BSA and possible solutions

Electronic records cannot be denied admissibility solely on the basis of being digital; therefore, deepfakes have to be kept under the purview of section 61 and section 62 of the BSA, thereby having a formal electronic evidence framework. As the initial burden is with the party submitting evidence, under section 63(4) an electronic evidence certificate along with the SHA-256 hash value, the unique cryptographic fingerprint, is submitted. If the file is altered post-creation, the hash value changes, making it easy for trial courts to address the concern, as it gives an immediate and non- forensic check against tampering.

As the Initial burden of proof lies on the party which submits the electronic evidence according to section 107, submission of the document under section 63(4) and the SHA-256 hash value satisfies the baseline requirement. Moreover, according to Section 106 of the BSA, the burden of proof as to any particular fact lies on the person who asserts it, and he cannot simply say that the evidence submitted by the proponent is forged; instead, he must submit prima facie evidence, for example, facial warping, visual artifacts, preliminary algorithmic analysis etc., to shift the burden of proof back to the proponent. The forensic investigation, as it does not have to be used as a routine course, can be ordered under section 39 (2) of the BSA, which governs the expert opinions of the examiners of electronic evidence notified under section 79A of the IT Act, once the challenger satisfies the section 106 burden. Operationalising the provisions of BSA requires balancing the technical standards and statutory requirements. Implementing the C2PA framework (Coalition for Content Provenance and Authenticity) enables verification of media origin and edit history; additionally, section 63(4) BSA certificates and SHA-256 hashes ensure post-seizure integrity. Therefore, SHA -256 cryptographic hash enables post-seizure integrity (that the file has not been tampered with post-seizure) but is unable to prove the authenticity of the original evidence, whereas the C2PA framework deals with the authenticity of the creation origin of evidence. By balancing the preliminary admissibility under section 107 and specific challenge thresholds enshrined under section 106, it ensures that the court directs the forensic evidence under section 39 (2) BSA only for genuine cases rather than requiring it in each and every case.

Conclusion and the Way Forward

To conclude, the deepfake menace is central with respect to the reliability of evidence in an AI-mediated world. Given the advantages deepfakes possess, the viable option will be to regulate their misuse. The need of the hour is to adopt the proportionality principle, in which authentication responsibility is on the party asserting the evidence and requires a genuine challenge before triggering the enhanced scrutiny and reserving the forensic investigation for rare cases that are in ardent need of it. India does not need to copy American, Canadian or Chinese legal provisions, but should strive to build its own legal framework where its own constitutional values are imbibed, to deal with the menace of AI-Generated deepfake content and thus bolster the trust of citizens in the Indian legal framework where justice must not only be done, but also it must seem to be done. As India does not have a dedicated single statute for dealing with this, codification of the laws dealing with this domain is a need of the hour.

---

[i] Daniel W. Linna Jr. et al., Deepfakes in Court: How Judges Can Proactively Manage Alleged AI-Generated Material in National Security Cases 4 (Aug. 8, 2024) (unpublished manuscript), https://ssrn.com/abstract=4943841.