AI Deepfakes as Evidence: Why Section 63 Certification Does Not Prove Content Authenticity
Summary: Section 63 of the Bharatiya Sakshya Adhiniyam, 2023 addresses the admissibility and procedural integrity of electronic records, but it does not establish that the underlying audio, video or image is genuine. A properly preserved deepfake can satisfy the procedural requirements of Section 63 while still depicting or recording an event that never happened. The Information Technology Act, 2000 addresses criminal and intermediary-side consequences through provisions including Sections 66D, 66E and 79, while the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 introduce a regulatory framework for synthetically generated information, including disclosure, labelling and compressed takedown timelines. Indian courts have addressed AI manipulation more directly in personality-rights cases such as Anil Kapoor v. Simply Life India & Ors. and the proceeding involving Amitabh Bachchan, but the evidentiary question of proving or challenging AI-manipulated content remains distinct. Anvar P.V. v. P.K. Basheer and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal remain key reference points for electronic-record certification, although neither was decided in the context of content that was synthetic from the moment of creation. In practice, parties challenging authenticity may require digital forensic examination of compression artefacts, metadata, voice biometrics and frame-level inconsistencies. Early preservation of the original file and metadata, particularised authenticity objections, forensic examination and attention to platform-level provenance information are therefore increasingly important.
Introduction
A company relies on a recorded video call to prove that a vendor agreed to revised commercial terms. The recording is properly preserved, the hash values match, and a Section 63 certificate is in place confirming the device and process used to generate the copy. The vendor’s counsel does not dispute any of this. Instead, they raise a different objection: the voice on the recording, they say, has been altered using AI tools, and the visuals do not match the vendor’s actual expressions during the call. The certificate proves where the file came from. It says nothing about whether what is in the file actually happened.
This is not a hypothetical problem anymore. It is the next stage of a question this series has been building toward since the first article on WhatsApp admissibility: once a court is satisfied that an electronic record is what it claims to be, in the sense of being an unaltered copy from a particular device, does that settle whether the content itself is genuine? For most records, the two questions collapse into one. For AI-manipulated content, they do not.
The Legal Issue
Indian evidence law, both under the old Indian Evidence Act, 1872 and now under the Bharatiya Sakshya Adhiniyam, 2023, was built around a specific kind of doubt: was this electronic record tampered with after it was created, and can its source and integrity be verified through a reliable process. Section 63 of the BSA answers exactly that doubt. It asks whether the computer or device that produced the record was functioning properly, whether the record accurately reproduces the information fed into or derived from the source, and whether a person familiar with the process can certify this.
None of that framework was designed to answer a different question: was the underlying audio, video or image itself fabricated at the point of creation, using tools that make fabricated content indistinguishable from genuine recordings. A deepfake video can pass every procedural test in Section 63 and still not depict anything that actually happened. The certificate authenticates the container. It does not authenticate the contents.
This distinction matters more now than it did even two years ago, because generative AI tools have made convincing audio and video forgery accessible without specialised skill, and because litigation increasingly involves recorded calls, voice notes and video evidence as primary proof in commercial disputes, employment matters and personal injury or matrimonial cases.
Applicable Legal Framework
Three separate legal frameworks bear on this issue, and they do different work.
Bharatiya Sakshya Adhiniyam, 2023, Section 63 continues to govern admissibility of electronic records as secondary evidence, requiring a certificate addressing the device, the process, and the person responsible for it. It remains the starting point for any electronic record placed before a court, deepfake or not, because a record that fails Section 63 does not even reach the question of content authenticity. The Bharatiya Sakshya Adhiniyam, 2023 contains the statutory framework governing electronic and digital records.
Information Technology Act, 2000 provides the criminal and platform-side response to deepfakes rather than an evidentiary one. Section 66D penalises cheating by personation using a computer resource, which covers impersonation carried out through synthetic audio or video. Section 66E addresses violation of privacy involving capture, publication or transmission of images of a person’s private area without consent, relevant where deepfake content is sexual or intimate in nature. Section 79 governs intermediary safe harbour, and this is the provision most directly reshaped by recent rule changes.
Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, notified by the Ministry of Electronics and Information Technology on 10 February 2026 (G.S.R. 120(E)) and in force from 20 February 2026, is the most significant recent development. For the first time, Indian law defines “synthetically generated information” as audio, visual or audio-visual content created or altered algorithmically so as to appear indistinguishable from a real person or event. The amendment requires platforms to seek disclosure from users uploading such content, mandates prominent labelling and, per subsequent draft amendments circulated for comment in 2026, continuous on-screen labelling for the duration of visual content rather than a one-time watermark. It also compresses takedown timelines sharply: three hours for unlawful content flagged by a court or government authority, and two hours for non-consensual deepfake content, down from the earlier 24 to 36 hour window. Failure to comply puts an intermediary’s Section 79 safe harbour at risk. The TaxGuru article on Government Proposes Tougher Information Technology Rules for Intermediaries discusses subsequent 2026 proposed amendments to the IT Rules, 2021.
It is worth being precise about what this framework does and does not do. It is a content moderation and platform compliance regime. It does not, on its own, create an evidentiary presumption that labelled content is synthetic or that unlabelled content is genuine, and no court has yet ruled on what weight, if any, a provenance label carries in a contested evidentiary proceeding. That gap is where litigants and counsel currently have to improvise.
Judicial Position
Indian courts have moved faster on the personality rights dimension of deepfakes than on the evidentiary dimension. In Anil Kapoor v. Simply Life India & Ors., CS(COMM) 652/2023, the Delhi High Court granted the actor an ex-parte omnibus injunction restraining unauthorised commercial use of his name, voice, image, likeness and persona, explicitly extending protection against AI-based tools such as face morphing and deepfake generation used without consent. A similar ad-interim order was granted to Amitabh Bachchan by the Delhi High Court in a 2022 proceeding restraining unauthorised commercial use of his voice, image and likeness, including through digitally manipulated content. These orders establish that using someone’s likeness through AI tools without consent can be restrained as a violation of personality and privacy rights.
That is a different question from admissibility. Neither order was decided in the context of a party trying to rely on, or challenge, AI-manipulated content as evidence of a fact in a civil or criminal proceeding. On that specific question, the settled reference points remain Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473, which made Section 65B certification (now Section 63) mandatory for secondary electronic evidence, and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1, which clarified when the certificate requirement can be dispensed with. Both judgments were written for a world where the central risk was an altered copy of a genuine recording, not a recording that was synthetic from the moment of creation. Neither addresses forensic authentication of content itself, and until a higher court does, trial courts are left to apply general principles of expert evidence and cross-examination to fill the gap.
Practical Examples
Consider three situations that are becoming increasingly common in practice. A former employee produces a voice recording purportedly of a manager making a discriminatory remark, and the employer’s counsel wants to know whether it can be challenged as fabricated rather than merely as inadmissible. A litigant in a matrimonial dispute submits a video allegedly showing a spouse’s conduct, and the other side suspects the video has been altered using widely available AI editing tools. A vendor disputes a recorded call used to prove contractual assent, on the ground that the audio has been synthetically generated or edited to insert words never spoken.
In each of these situations, a Section 63 certificate addressing the device and storage process will not resolve the dispute, because the challenge is not to the chain of custody but to the authenticity of the content itself. That requires forensic examination distinct from what the certificate covers: analysis of compression artefacts, metadata inconsistencies, voice biometric comparison, or frame-level inconsistencies that specialised digital forensic examiners, rather than the certifying witness under Section 63, are equipped to detect.
Common Mistakes
Litigants and counsel frequently treat a properly certified electronic record as beyond challenge once Section 63 formalities are satisfied. That is a mistake, because certification addresses provenance, not content genuineness, and an opposing party is fully entitled to challenge authenticity separately through cross-examination and forensic evidence.
A related mistake is assuming that the absence of a synthetic content label under the 2026 IT Rules means a recording is genuine. The labelling regime is a platform compliance obligation aimed at reducing public harm from unlabelled deepfakes circulating online. It was not designed as, and does not function as, a courtroom authentication mechanism, and unlabelled content can still be synthetic if it was never uploaded through a platform covered by the rules, or if detection and disclosure obligations were not triggered or complied with.
A third mistake, more practical than legal, is failing to preserve the original file and its full metadata at the earliest possible stage. Once content is compressed, re-recorded, or passed through messaging platforms that strip metadata, the forensic markers that would help establish or rule out manipulation are often lost, which weakens the position of a party who genuinely holds a real, unaltered recording.
Practical Steps
Where a party intends to rely on audio-visual electronic evidence in a proceeding where authenticity is likely to be contested, early forensic preservation of the original file, including full metadata, should happen before the file is shared, compressed, or transmitted through third-party platforms. Where the opposing side raises a synthetic content objection, that objection should be tested through a qualified digital forensic expert rather than left to inference from surrounding circumstances alone. Counsel should also track the platform-level record: whether the platform on which the content originated or circulated was subject to the 2026 labelling and disclosure obligations, and whether any label, provenance metadata, or disclosure was present or absent, since this can support, though not conclusively resolve, an authenticity argument. Finally, given that this remains an evolving area without settled appellate guidance, drafting pleadings and evidence affidavits with specific, particularised authenticity objections, rather than generic denials, will matter more in getting a court to actually engage with the forensic question rather than treating it as a routine admissibility dispute.
Conclusion
The digital evidence framework built through Section 63 and the case law around it was designed to answer whether a copy is a faithful copy. It was not designed to answer whether the original was ever real. As AI-generated content becomes harder to distinguish from genuine recordings, that gap is where contested litigation is increasingly likely to move, and the 2026 IT Rules, while a meaningful regulatory step for platform accountability, leave the courtroom question largely open. Practitioners who treat certification and content authenticity as the same inquiry will find themselves unprepared for objections that are becoming routine rather than exceptional.
******
Disclaimer: This article is intended for general legal awareness and information purposes only. It does not constitute legal advice and should not be relied upon as a substitute for consultation with a qualified legal professional in relation to specific facts and circumstances. While reasonable care has been taken to ensure accuracy as of the date of publication, readers are advised to independently verify the current statutory and judicial position before relying on it, particularly given that the regulatory framework discussed is still evolving.
