Inside the SC’s proposed regulations on the use of AI in courts: What’s allowed and what’s absolutely prohibited | News Explained

What role has the SC given AI?

The draft regulations propose requiring courts to “actively seek opportunities” to introduce AI systems and tools that “obviously” improve access to justice, reduce delays and increase administrative efficiency.

The use of AI is explicitly permitted for a variety of administrative and support functions, including case management, transcription, translation, legal research, document summarization, accessibility, and court operations. All these require written approval by the Supreme Committee of the SC or the AI ​​Committee of the High Court/Court, as the case may be, and the supervision and confirmation of an officer appointed by the court.

Can AI judge incidents?

No, the draft regulation clearly provides that no judicial outcome can be based solely on algorithmic decision-making or AI-generated information. Human jurisdiction determines all judicial decisions. Therefore, the use of AI in decision-making processes related to cases is advisory only and subject to independent human (judicial) evaluation.

What is off limits?

Some uses are prohibited on “absolute and non-derogable” terms, meaning that they cannot be later authorized by any authority under the regulations. These include ‘risk scoring’ to assess flight risks. Predicting recidivism. Assessing bail eligibility; determining the credibility of witnesses. Use AI to predict, profile, or infer the future actions or actions of parties, defendants, witnesses, or legal representatives more generally. Submit AI-generated output as independent evidence without fully disclosing the nature of the AI-generated output. Using black box (unaccountable) AI systems for issues that affect personal freedom.

Will litigants be able to know if AI was used in their case?

Yes, the draft regulations propose that when a court uses an AI tool to “substantially assist” in case management, document analysis, or any aspect of judicial administration, it must notify the parties (litigants and their attorneys) in a timely and accessible manner. The standard is material aid. Therefore, litigants will not be informed of every instance in which AI has been used in their case, but only if the AI ​​has provided substantial assistance to the judge.

Who will regulate it?

The organizational structure will be led at the highest level by the SC’s “apex body,” which will be comprised of sitting SC and high court judges, Ministry of Electronics and Information (MeitY) officials, and financial and cybersecurity experts. This agency establishes minimum mandatory standards that approve the use of AI systems and issues implementation guidelines. It will work through five specialized committees. The SC and each High Court will constitute its own AI Committee with the support of the AI ​​Secretariat.

A separate research organization, the Center for Research Excellence in Artificial Intelligence (CoRE-AI), evaluates tools and tracks development in support of the Apex Body.

Will private companies be involved?

yes. However, written approval from the relevant court authority is required and a mandatory list of terms and conditions must be followed. Vendor agreements should include ownership of and access to forensic data and AI output, prohibitions on the use of sensitive judicial data, and an explicit prohibition on retaining or fine-tuning models using forensic data without written approval from the AI ​​committee.

The draft regulations also provide that vendors cannot claim proprietary intellectual property over tools that are substantially built on judicial data or public resources.

What safety measures are in place?

The draft regulations take a lifecycle approach to safety measures before, during, and after the deployment of AI systems. A technical and ethical impact assessment will be required covering system architecture, training data, bias, hallucination risks, and cybersecurity posture. Some systems may require you to first perform “controlled environment testing” in an isolated setup. After implementation, the system will undergo technical, legal, and ethical audits, as well as separate cybersecurity audits during the same cycle. These are conducted in-house, as source code and training data cannot be shared with third parties for audit purposes.

Each court would also be required to maintain an AI registry documenting approved systems and audit results, and each AI office would be required to maintain an AI incident database to record malfunctions, errors, and biases, and to notify within 24 hours if a tool fails or is shut down. Each high court will also be required to maintain emergency fallback protocols to keep court processes running manually even if the system goes down.

What recourse does litigants have if AI used in litigation causes harm?

The complaints route applies in particular where harm is caused by the use of prohibited AI (as explained above). Affected parties may file an application at the earliest opportunity with the court in which the relevant AI system is or will be used. The court must give hearing to the affected parties and make “such appropriate orders as it deems fit.”

The High Court may establish a separate simplified procedure and form of complaint to make it accessible to people with limited legal knowledge. This grievance redressal stream runs parallel to other legal remedies already available under regular law.

The author is a Research Fellow and Senior Resident Fellow in the Center for Applied Law and Research at the Vidi Center for Law and Policy.