Kerala Mandates AI for Witness Depositions, Becoming First State to Digitize Evidence Recording Statewide

The Indian judiciary, traditionally characterized by manual processes and paper-based evidence, is on the cusp of a systemic transformation. The catalyst for this change is the Kerala High Court, which recently issued an official memorandum mandating the adoption of 'Adalat.AI', an Artificial Intelligence-based speech-to-text transcription tool, for recording all witness depositions in district courts across the state. This directive, set to take effect from November 1, 2025, marks a historic moment: Kerala is the first state in India to make AI transcription mandatory across its entire subordinate judicial system.

The move, spearheaded by the High Court's Directorate of IT, follows a successful pilot phase initiated earlier this year in select courts within the Ernakulam district. The core objective is clear: to modernise court processes, minimize delays in recording evidence, and significantly enhance transparency and accuracy in judicial proceedings, thereby addressing the persistent issue of judicial backlog.

Until now, the process of recording testimony relied heavily on the presiding judge handwriting the statement or dictating it to a court typist—a time-consuming and error-prone exercise that frequently led to adjournments and delays. Judicial officers, facing a severe shortage of dedicated court stenographers, often cited the manual documentation of evidence as a primary bottleneck in trial progression. The High Court memorandum directly addresses this inefficiency, stating that the primary mode of recording depositions from the appointed date will be the Adalat.AI Voice-to-Text Transcription Tool.

Automating the Crux of Justice: Speed vs. Fidelity

The integration of Adalat.AI aims to automate what is arguably the most crucial phase of a trial: the recording of oral evidence. The technology, developed by a deep-tech startup and trained on a specialised legal speech model, supports both Malayalam and English and is designed to recognize and transcribe complex legal jargon and regional phrases with high accuracy.

The claimed operational benefits are transformative:

• Real-Time Transcription: The instantaneous conversion of spoken testimony into text eliminates the lag associated with manual writing or dictation, allowing trials to proceed more smoothly and quickly.

• Reduced Adjournments: By removing the dependency on human speed for transcription, the system is expected to drastically cut down the number of hearings adjourned solely for the purpose of completing the recording of evidence.

• Enhanced Accessibility: Once recorded, endorsed, and signed, the depositions will be uploaded to the District Court Case Management System (DCMS), providing lawyers and parties with immediate, digital access to the certified documents through their respective dashboards.

• Improved Accuracy: The consistent, machine-generated record reduces the potential for human error, ambiguity, or subjective interpretation inherent in handwritten or dictated summaries of testimony. Furthermore, a screen will be provided in the witness box, allowing the witness to review their statement digitally before providing their endorsement and signature, solidifying the record’s integrity.

This leap aligns perfectly with the broader national push under the e-Courts Mission Mode Project (Phase III), which seeks to integrate new technologies, including AI, to create a faster, more accessible, and transparent judicial system across India. Kerala’s mandatory adoption of AI sets a robust and ambitious precedent for other High Courts considering digital reforms.

The Evidentiary Challenge of Algorithmic Transcription

While the efficiency gains are undeniable, the mandatory deployment of AI in recording evidence—the bedrock of judicial fact-finding—raises critical legal and procedural questions that must be addressed rigorously.

The fundamental legal framework governing evidence recording is enshrined in the Indian Evidence Act, 1872, and the procedural codes, namely the Code of Civil Procedure (CPC) and the Code of Criminal Procedure (CrPC). These statutes place immense weight on the authenticity and veracity of the record of testimony. Specifically, the deposition (or statement) of a witness forms part of the judicial record and serves as the primary material for appellate review and for confronting the witness with prior inconsistent statements under Section 145 of the Evidence Act.

The introduction of AI transcription necessitates a fresh look at the concept of the "judicial record":

1. Authentication and Endorsement: While the memorandum explicitly requires the deposition to be endorsed and signed, the presiding judge’s ultimate role remains paramount. The judge must confirm that the algorithmic output accurately reflects the oral testimony. This human layer of verification is essential to maintain the integrity of due process and prevent the process from being challenged on the grounds of flawed transcription.

2. Handling Errors and 'Hallucinations': No AI system is infallible. Errors arising from poor courtroom acoustics, overlapping speech, distinct regional accents, or linguistic variations (even within Malayalam or English) could result in inaccurate or misleading transcripts—a phenomenon sometimes referred to as 'AI hallucination.' Who bears the legal liability if a material omission or inaccuracy in the AI-generated record compromises the outcome of the trial? The High Court's prior "Policy Regarding Use of Artificial Intelligence Tools in District Judiciary" (July 2025) likely addresses this by mandating human verification of AI outputs, ensuring judges retain full responsibility for the content and integrity of any judicial record.

3. The Nuance of Testimony: Court testimony is not merely a string of words; it includes non-verbal cues, inflections, pauses, and the witness's demeanour, which the presiding officer conventionally captures through descriptive notes. AI transcription captures the literal content but may obscure the contextual nuances vital for assessing credibility. Future iterations of the system must address the need to integrate audio-visual records alongside text to preserve the holistic quality of the evidence.

Data Sovereignty and Security Protocols

The shift from physical paper records held locally to digital transcripts uploaded to the DCMS—and potentially to a cloud system—involves complex data security and privacy implications. Witness statements, particularly in sensitive cases (such as those under the POCSO Act or involving sexual violence against women and children), constitute highly confidential Personal Data under the Digital Personal Data Protection Act, 2023 (DPDP Act).

The Kerala High Court has demonstrated awareness of this by strictly controlling technology integration. The memorandum stipulates that in case of technical difficulty with Adalat.AI, courts may only use alternative platforms duly approved by the Directorate of IT, High Court, ensuring data security and confidentiality. This central vetting mechanism suggests a commitment to Data Sovereignty and ensuring that sensitive judicial data remains within a protected framework.

The mandatory nature of the system dictates that robust protocols must cover:

• Access Control: Strict authentication mechanisms to ensure only authorised judicial officers, court staff, and registered advocates can access the deposition transcripts via their dashboards.

• Tamper-Proof Records: Cryptographic measures (like digital signatures and timestamping) to guarantee that once a transcript is endorsed and signed, it cannot be altered without detection, thus maintaining its evidentiary sanctity.

• Infrastructure Equity: The rollout must ensure that all district courts, especially those in rural or less-connected regions, have the necessary hardware, stable high-speed internet, and reliable power backup to support mandatory real-time AI usage.

The Path Forward: A National Blueprint

Kerala’s decision transcends mere technological adoption; it represents a bold institutional decision to change the fundamental way evidence is captured in court. By mandating the use of Adalat.AI, the High Court is moving the burden of transcribing from the overloaded judicial officer to the machine, allowing judges to focus on the substantive task of judicial application of mind, which involves critical listening, observation of witness demeanour, and assessing the evidentiary value of the testimony.

This initiative sets a national benchmark for legal technology deployment, particularly because of its mandatory, statewide scope and its focus on the primary evidence-recording stage.

If successfully implemented, the Kerala model could serve as a blueprint for the national judiciary, proving that AI can be a transformative force in combating judicial delays. The success of this immense undertaking, however, hinges not just on the accuracy of the software but on three crucial factors: comprehensive training for all judicial stakeholders, the development of robust, fail-safe protocols for technical malfunctions, and a continuous feedback loop to refine the AI's language models against the diverse linguistic landscape of the state.

The judicial ecosystem is keenly watching Kerala. The mandatory integration of Adalat.AI is poised to test the adaptability of India's judicial procedure against the swift power of deep technology, promising a future where justice is not only accurate but also delivered in a timely fashion.