Deepfakes as Evidence: Evidentiary Admissibility Standards in an Era of Synthetic Media
Introduction
In October 2024, a criminal trial in a Sessions Court in Mumbai encountered an unprecedented evidentiary challenge. The prosecution’s key exhibit was a video recording purportedly showing the accused at the scene of the offence. The defence produced an expert witness who testified that the video had markers consistent with AI-generated facial substitution, commonly referred to as a deepfake. The court, facing no established procedure for evaluating such testimony, adjourned the matter three times while seeking guidance from the High Court on admissibility. The High Court, in turn, found no binding precedent.
This episode is not isolated. Across Indian courts, evidence in digital form is increasingly being challenged on the grounds of AI-generated manipulation. The deepfake problem, which once seemed a concern limited to political disinformation and celebrity image abuse, has entered the courtroom. Video evidence of crimes, audio recordings of conversations, digital photographs of documents, and even purported CCTV footage are all potentially subject to synthetic manipulation that is, at current levels of generation technology, undetectable by the human eye.
The Indian evidence law framework, considerably modernised by the Bharatiya Sakshya Adhiniyam, 2023, retains conceptual gaps that deepfake evidence exposes with particular clarity. Without a dedicated standard for authenticating digital evidence against the risk of AI manipulation, courts face a binary and unsatisfactory choice: admit all digital evidence and risk convicting on fabricated material, or exclude contested digital evidence and risk acquitting the guilty.
Legal Framework
The Bharatiya Sakshya Adhiniyam, 2023, which replaced the Indian Evidence Act, 1872, with effect from July 2024, modernises evidentiary law in several respects. Section 63, providing for computer outputs as documentary evidence, requires a certificate from a responsible official of the organisation that produced the electronic record, confirming the integrity and reliability of the computer system and the output.
The certificate requirement under Section 63 was interpreted by the Supreme Court in Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal (2020) to be mandatory for admissibility. This authentication framework, designed for the era of computers producing records of human transactions, is wholly inadequate for the era of AI-generated media, where the relevant question is not whether a computer produced the record but whether the AI was manipulated to produce a false one.
The Bharatiya Nyaya Sanhita, 2023, contains provisions on forgery and the use of false evidence. A deepfake introduced into court proceedings would constitute fabricated false evidence under these provisions. However, these provisions address the act of the party who introduces the evidence, not the standard of authentication that courts must apply to all digital evidence.
Judicial Developments
The Karnataka High Court, in a 2024 bail matter, encountered a challenge to audio evidence on deepfake grounds for the first time on the Indian record. The court directed that a forensic examination by the Central Forensic Science Laboratory be conducted before the audio could be relied upon. The CFSL report identified certain spectral anomalies consistent with AI-generated synthesis but was unable to definitively conclude manipulation. The court admitted the audio provisionally, noting its weight would be assessed at trial.
Germany’s Constitutional Court, in a 2025 preliminary injunction challenging algorithmic risk assessment in preventive detention, held that a deprivation of liberty premised on future risk prediction must satisfy the strictest standard of justification, including a right to disclosure of the algorithm’s methodology and an opportunity for the affected individual to contest the factual inputs on which the risk score was based.
Contemporary Issues and Analysis
The core evidentiary challenge posed by deepfakes is not detection, though detection is genuinely difficult. It is the allocation of the burden of proving authenticity. The forensic detection of deepfakes relies on a combination of pixel-level analysis, compression artefact examination, temporal consistency analysis, and biological signal detection. As of 2025, the best detection algorithms achieve accuracy rates between 85 and 93 percent on benchmark datasets, but their performance on real-world samples produced by state-of-the-art generation models is significantly lower.
This probabilistic character creates a fundamental tension with the standard of proof in criminal proceedings. Beyond reasonable doubt requires a court to be morally certain of the accused’s guilt. If the authenticity of a key piece of video evidence is established at a 90 percent probability of being genuine, the 10 percent probability that it is synthetic deepfake media may itself constitute reasonable doubt.
There is also the institutional capacity problem. The CFSL and state forensic science laboratories in India have limited capacity for deepfake forensic examination. Court-appointed forensic examiners with specific deepfake detection expertise are extremely rare.
Comparative and International Perspective
The United States Federal Rules of Evidence require electronic evidence to be authenticated by evidence sufficient to support a finding that the item is what the proponent claims it is. Federal courts have begun requiring proponents of video evidence to produce both the device that captured the recording and a chain of custody log establishing its integrity from capture to court presentation.
The Uniform Law Commission has begun work on a Model Act on Synthetic Media Evidence, expected to be finalised in 2026. The United Kingdom’s Law Commission’s 2025 review of digital evidence standards has recommended that English courts adopt a reliability rather than authenticity standard for electronic evidence.
Practical and Policy Implications
For criminal justice, the deepfake evidence problem has consequences in both directions. If courts are too credulous toward digital evidence, fabricated media can secure wrongful convictions. If courts are too sceptical, legitimate digital evidence may be excluded on speculative deepfake grounds, enabling impunity. The calibration must be precise, and it requires judicial education, institutional forensic capacity, and clear evidentiary rules.
For civil litigation, particularly in defamation, employment, and family law matters, deepfake evidence challenges are already being used tactically. A party who cannot afford the cost of genuine deepfake forensic analysis is at a disadvantage in challenging manipulated media introduced by a better-resourced opponent.
Suggestions and Reforms
India should amend the Bharatiya Sakshya Adhiniyam to introduce a specific provision on AI-manipulated evidence, establishing that any party who relies on video, audio, or image evidence that is challenged on synthetic manipulation grounds must produce a forensic authentication certificate from an accredited laboratory. The forensic certificate should specify the methodology used and the confidence level of the authentication finding.
The National Forensic Science University and the CFSL should be mandated and resourced to develop a certified deepfake detection laboratory with standardised protocols and rapid-response capability for urgent court-directed examinations. A judicial training programme on AI evidence, developed jointly by the National Judicial Academy and technical experts, should be delivered to judges across all court levels.
Conclusion
Evidence is the foundation of justice. When that foundation can be synthesised with tools accessible to ordinary litigants, and when the legal system has no reliable method to distinguish genuine from fabricated, the administration of justice is in genuine danger. India’s evidence law reforms, recent and welcome as they are, did not anticipate this challenge. Addressing it now, before deepfake-related miscarriages of justice accumulate, is a matter of institutional urgency that courts, legislators, and forensic institutions must approach together.
