AI-generated evidence in Indian criminal trials

The Ghost in the Machine: Navigating AI-Generated Evidence in Indian Criminal Trials

The quiet tap of a keyboard in a forensic lab now carries as much weight as a witness’s testimony in an Indian courtroom. As artificial intelligence (AI) evolves from a conceptual marvel into a daily investigative tool, the Indian judiciary faces a transformative challenge. From deepfake detection and facial recognition to predictive profiling and AI-assisted forensic reconstruction, “algorithmic evidence” is no longer science fiction. It is a reality that tests the boundaries of the Bharatiya Nyaya Sanhita (BNS) and the Bharatiya Sakshya Adhiniyam (BSA).

The transition from the colonial-era Indian Evidence Act 1872 to the BSA 2023 was intended to modernize the legal framework for the digital age. However, as AI systems transition from “tools” to “decision-makers,” the question remains: is the Indian legal system equipped to cross-examine a black box?

1. The Digital Metamorphosis: Defining AI Evidence

In the context of Indian criminal law, AI-generated evidence typically falls into two categories: primary digital evidence (e.g., a deepfake video intended to frame an accused) and derivative analytical evidence (e.g., AI-enhanced CCTV footage or GAN-based reconstructions).

Under Section 2(1)(i) of the BSA, the definition of a “document” has been expanded to include electronic records. While this captures the storage of AI data, it does not fully address the generation of it. Unlike traditional digital evidence, which is a static snapshot of a past event, AI-generated evidence is often probabilistic. When an AI tool enhances a blurred license plate, it is not “restoring” pixels; it is “predicting” what those pixels should look like based on a trained dataset.

2. Theoretical Foundations and the BSA 2023

The admissibility of AI evidence hinges on Section 63 of the BSA (formerly Section 65B of the Evidence Act), which governs the admissibility of electronic records. The requirement for a certificate to ensure the integrity of the source remains a cornerstone of the process.

2.1 The Problem of Authenticity

In Anvar P.V. v P.K. Basheer, the Supreme Court emphasized that electronic records are fragile and easily manipulated. AI intensifies this fragility. Generative Adversarial Networks (GANs) can now create content that bypasses the “human eye” test. If a defense counsel alleges that a video confession is a deepfake, the burden of proof under Section 60 of the BSA—which demands direct oral evidence—becomes complicated. If the “source” is an algorithm, who provides the direct evidence?

2.2 Expert Opinion under Section 39

Section 39 of the BSA allows for the opinion of examiners of electronic evidence. However, there is a burgeoning “competency gap.” A forensic expert might understand hardware, but the “black box” nature of proprietary AI algorithms makes it nearly impossible for an expert to explain why an AI reached a specific conclusion. This lacks the transparency required for a “fair trial” under Article 21 of the Constitution of India.

 

3. The “Black Box” Challenge and Due Process

One of the most human elements of a criminal trial is the right to confront one’s accuser. When an AI system flags a suspect via predictive policing or facial recognition, the “accuser” is a set of weighted nodes in a neural network.

3.1 Transparency vs. Proprietary Rights

Many AI tools utilized by law enforcement agencies are developed by private corporations. These companies often protect their algorithms as trade secrets. In a criminal trial, if the defense cannot audit the code for bias—such as racial or gender biases in facial recognition—the right to a robust defense is compromised. The Indian judiciary must decide whether a developer’s intellectual property outweighs an accused person’s liberty.

3.2 The Risk of “Automation Bias”

Judges and police officers are not immune to “automation bias”—the tendency to favor suggestions from automated systems over human intuition. In Indian trials, where the burden of proof is “beyond reasonable doubt,” the veneer of mathematical certainty provided by AI can be dangerously seductive.

4. Deepfakes: The New Frontier of Forgery

The rise of synthetic media poses a dual threat to Indian criminal justice. On one hand, deepfakes can be used to create false evidence; on the other, the mere existence of deepfakes allows guilty parties to claim that authentic evidence is “AI-generated.” This is known as the “liar’s dividend.”

To combat this, Indian courts must move toward a “Chain of Custody for Pixels.” This involves:

1. Metadata Analysis: Verifying the temporal and spatial markers of digital files.

2. AI-Transparency Logs: Requiring forensic labs to provide the “training parameters” of any AI tool used to analyze evidence.

5. Judicial Precedents and the Road Ahead

While the Indian Supreme Court hasn’t yet issued a definitive “AI Evidence Code,” its trajectory in cases like Arjun Panditrao Khotkar v Kailash Kushansrao Gorantyal suggests a strict adherence to procedural safeguards for digital data.

In international jurisdictions, we see the emergence of the “Daubert Standard”—a rule of evidence regarding the admissibility of expert witness testimony. India needs a localized version of this: a “Techno-Legal Standard” that evaluates whether the AI methodology has been peer-reviewed and what its known error rate is within the Indian demographic context.

6. Recommendations for a Human-Centric Legal Framework

To ensure that AI serves justice rather than subverting it, the following reforms are essential:

6.1 Legislative Clarity

The BSA should be amended to include a specific category for “Algorithmic Records.” This would distinguish between a simple PDF and a probabilistic output from an AI model.

6.2 The “Human-in-the-Loop” Mandate

No conviction should be based solely on AI-generated analytics. AI should be treated as “corroborative evidence” rather than “substantive evidence.” The final interpretive leap must always be performed by a human forensic expert who can be cross-examined.

6.3 Specialized Digital Courts

The complexity of AI requires a judiciary that understands the nuances of data science. Training programs for judicial officers should focus on “algorithmic literacy,” enabling them to identify when an AI tool is being over-relied upon.

7. Conclusion: Balancing Innovation and Liberty

AI-generated evidence in Indian criminal trials is a double-edged sword. It offers the potential to solve cold cases and bring unprecedented precision to forensic science. However, without stringent safeguards, it risks introducing “black box” injustices into a system already burdened by delays.

As we navigate the era of the Bharatiya Sakshya Adhiniyam, the focus must remain on the human element. Technology may provide the “what,” but the law must always ask “why” and “how.” The ghost in the machine must be unmasked before it is allowed to take the witness stand.

Table of Authorities

Statutes

• Bharatiya Sakshya Adhiniyam 2023 (BSA)

• Bharatiya Nyaya Sanhita 2023 (BNS)

• Constitution of India 1950

• Information Technology Act 2000

Cases

• Anvar P.V. v P.K. Basheer (2014) 10 SCC 473

• Arjun Panditrao Khotkar v Kailash Kushansrao Gorantyal (2020) 7 SCC 1

• Shafhi Mohammad v State of Himachal Pradesh (2018) 2 SCC 801

Bibliography

Books

• Sarkar S, Sarkar’s Law of Evidence (19th edn, LexisNexis 2016)

• Reed C, Making Laws for Cyberspace (OUP 2012)

Articles

• Dhananjaya Y. Chandrachud, ‘Technology and the Changing Face of Justice’ (2023) 5 Indian Law Review 1

• Surana A, ‘Admissibility of Digital Evidence in India’ (2021) 4(2) International Journal of Law Management & Humanities 450

Footnotes (OSCOLA Style Samples)

1. Bharatiya Sakshya Adhiniyam 2023, s 2(1)(i).

2. Anvar P.V. v P.K. Basheer (2014) 10 SCC 473.

3. Constitution of India 1950, art 21.

4. Arjun Panditrao Khotkar v Kailash Kushansrao Gorantyal (2020) 7 SCC 1.

5. S Sarkar, Sarkar’s Law of Evidence (19th edn, LexisNexis 2016) 1102.

6. Information Technology Act 2000, s 43A.