UNDP’s Digital Office leads the organization’s global work on Artificial Intelligence under an “AI for Development” mandate, focused on strengthening public institutions and advancing inclusive development through Responsible and Trusted AI governance that addresses bias, transparency, accountability, and human oversight. This approach supports responsible and scalable AI adoption across sectors including justice, health systems, agriculture, social protection, climate resilience, and public administration.

Access to justice and the rule of law are foundational to democratic governance, the protection of human rights, and inclusive and sustainable development. In recent years, India’s judicial ecosystem has increasingly turned to adopting artificial intelligence to support judicial administration and make justice delivery more efficient and responsive. However, the usage of AI within digital contexts also raises significant questions related to fairness, transparency, accountability, privacy, and the protection of fundamental rights.  As AI systems increasingly intersect with processes that affect legal outcomes and access to remedy, ensuring that their adoption is governed by clear, rights-based, and institutionally grounded frameworks becomes essential to preserving the integrity, legitimacy, and equity of the justice system.

In this context, the United Nations Development Programme (UNDP) in India, in partnership with DAKSH and the Digital Futures Lab, undertook the report AI for Justice: Ethical, Fair and Robust Adoption in India’s Courts. The report provides a comprehensive assessment of the emerging use of artificial intelligence within Indian courts, examining how AI tools are currently being introduced, the governance gaps that accompany their deployment, and the implications for judicial integrity, access to justice, and the protection of rights.

Key Insights:

• AI adoption in Indian courts is expanding but remains fragmented and insufficiently governed.
  AI tools are being introduced primarily through isolated pilots and experimental initiatives, often without clear documentation, defined objectives, or consistent evaluation mechanisms. This limits institutional visibility over where and how AI is being used and weakens accountability within the judicial ecosystem. 
• Efficiency-driven deployment frequently outpaces institutional readiness.
  Courts are adopting AI solutions to address caseload pressures and administrative inefficiencies without systematic assessment of their human, technical, and financial capacity to deploy, oversee, and sustain these systems responsibly over time. 
• Judicial contexts amplify rights-based risks associated with AI use.
  The deployment of AI in courts raises heightened concerns related to fairness, transparency, explainability, privacy, and due process, particularly given the sensitive nature of judicial data and the potential impact of AI outputs on legal outcomes and access to remedy.
• Governance and accountability frameworks remain underdeveloped.
  The absence of standardized approaches to risk assessment, vendor evaluation, human oversight, and post-deployment monitoring increases dependence on private technology providers and poses risks to judicial independence, public trust, and equitable justice delivery.

Read the full report here.

In response to the governance, rights-based, and institutional gaps identified in the assessment, the report proposes a structured, risk-based approach to guide the responsible adoption of artificial intelligence within the judicial ecosystem. These assessment frameworks recognize [SD1]  that judicial contexts demand heightened safeguards due to their direct implications for rights, due process, and public trust, while providing courts with practical tools to assess institutional readiness, identify and mitigate risks, evaluate technology providers, and ensure continuous oversight throughout the AI lifecycle.

Access the assessment frameworks here:

1.   Institutional Readiness Assessment: https://bit.ly/4b4ojX1

2.  Risk Assessment by Use-Case: https://bit.ly/3OTrLfN

3.  Technical Assessment for Solution Providers and Vendors: https://bit.ly/4smduqm

4.  Ongoing-Continuous Assessment: https://bit.ly/4stHmkS

The Four-Part Assessment Approach

• Institutional Readiness Assessment
  This framework supports courts in evaluating their preparedness to adopt AI responsibly. It examines governance structures, human and technical capacity, financial sustainability, and internal accountability mechanisms necessary to oversee AI systems effectively.
• Risk Assessment
  The risk assessment framework enables courts to identify potential harms associated with specific AI use cases, including risks to fairness, equality, privacy, and due process. It supports decision-making on whether a proposed application should proceed, require additional safeguards, or be deemed unsuitable for judicial use.
• Technical and Vendor Assessment
  This component provides a structured method for evaluating AI tools and technology providers, focusing on data governance, model transparency and explainability, robustness, cybersecurity, and compliance with legal and ethical standards.
• Ongoing and Continuous Assessment
  Recognizing that risks evolve over time, this framework supports post-deployment monitoring of AI systems. It emphasizes performance evaluation, impact assessment, documentation, grievance redress mechanisms, and periodic review to ensure sustained accountability.

How to use these frameworks:

These assessment tools are designed to function as a sequenced and documented decision-making process for courts considering AI adoption. Each workbook contains structured indicators, scoring logic, and summary outputs that must be completed in full before progressing to the next stage.

1. Institutional Readiness as a Precondition

Begin with the Institutional Readiness Assessment. Courts should respond to all essential questions using the designated dropdown and free-text fields, ensuring that supporting justification is recorded. The sheet auto-generates a cumulative score reflected in the “Summary” tab.

A minimum threshold score (≥60%) indicates sufficient baseline preparedness to proceed to use-case evaluation. Where this threshold is not met, identified limitations must be addressed before progressing further. This stage ensures that AI adoption is not pursued in the absence of governance structures, capacity, oversight mechanisms, and accountability safeguards.

2. Risk Classification at the Functional Use-Case Level

For each proposed AI application, complete the Risk Assessment by Use-Case. The tool requires classification of the system based on its function, level of autonomy, potential impact on rights, and affected stakeholders. The framework auto-generates a risk level:

• Low Risk
• Medium Risk
• High Risk
• High Risk – Recommending Prohibition

The classification determines whether adoption may proceed, whether mitigation measures are required, or whether the tool should not be deployed. Where risk outweighs potential benefit, courts are advised to refrain from use.

3. Technical and Vendor Scrutiny Proportionate to Risk

If the use-case proceeds, apply the Technical Assessment for Solution Providers. The level of scrutiny is calibrated to the risk classification.

• For Medium Risk (and, where appropriate, Low Risk) contexts, the shorter questionnaire may be used.
• For High Risk contexts, the extended questionnaire must be completed.

Courts should require documentation relating to data governance, model performance, explainability, cybersecurity, auditability, and accountability. For High Risk systems, progression is recommended only where a sufficiently high cumulative score (≥60%) is achieved. This stage addresses information asymmetries and ensures defensible procurement and deployment decisions.

4. Ongoing and Continuous Assessment

AI governance does not conclude at deployment. The Ongoing/Continuous Assessment framework establishes a periodic review cycle to monitor performance, document incidents, reassess risks, and update mitigation measures.

Prior to deployment, courts should inform vendors that access to baseline metrics, performance logs, model documentation, and evaluation benchmarks will be required. Where the framework indicates “Action Required,” corrective measures must be undertaken. If limitations are identified that materially affect safeguards, the tool should be paused until rectified.

For more details, please contact:
1. Nusrat Khan - nusrat.khan@undp.org
__________________________________________________________________________________________________________________________