New Delhi: The Supreme Court this week released draft regulations for the use of artificial intelligence (AI) in courts, placing “human primacy” at the centre of judicial AI governance and emphasising that AI must remain an assistive tool rather than a decision-maker.
As courts across the world grapple with the use of AI in courtrooms by litigants, lawyers and judges alike, India joins a host of other countries drawing up such rules on the use of artificial intelligence in courts.
The draft says AI systems can function only in an “assistive capacity” and cannot replace judicial officers in determining questions of law, fact or justice. Responsibility for any decision taken with AI assistance continues to rest exclusively with the concerned judicial officer.
The framework also requires disclosure of AI use in legal filings, mandates human supervision of AI-assisted processes, and proposes regular audits, training programmes and governance mechanisms to oversee judicial AI deployment.
How does India’s framework compare to those of other countries? Three broad models have emerged globally—the governance model, the guidance model and the smart-courts model. Each treats AI slightly differently.
ThePrint unpacks the three models and how they differ from each other.
Also Read: Employees are AI-ready, but organisations holding them back, says new Microsoft study
The EU model
The EU has taken a strong stance on AI regulation ever since the proliferation of generative models began. Unusually, its rules on AI court usage are not separate regulations, but folded directly into the 2024 AI Act, which gives them the force of law across EU member states.
In the EU framework, the use of AI in justice is considered a “high-risk” application, putting it in the same category as AI components in transport infrastructure and healthcare.
These systems are then subject to “strict obligations”. In practice, this means judicial AI systems must meet requirements on risk management, human oversight, transparency, record-keeping and auditability before they can be deployed.
As part of these requirements, all these systems must be overseen by humans who have been trained and are aware of potential AI-related issues. They must, in particular, “remain aware of the possible tendency of automatically relying or over-relying on the output produced by a high-risk AI system (‘automation bias’)”.
Those assigned oversight have full authority to disregard and override any AI outputs.
“The use of AI tools can support the decision-making power of judges or judicial independence, but should not replace it: the final decision-making must remain a human-driven activity,” the EU AI Act declares in its preamble.
Even for allowable uses, the AI Act demands strict accountability measures. AI usage must be automatically logged and subject to scrutiny to ensure “a level of traceability of the functioning of a high-risk AI system”, according to the regulations.
Further restrictions come from the EU’s strict data protection regime, the General Data Protection Regulation (GDPR), which imposes additional requirements for the lawful processing of personal data, transparency, purpose limitation and data minimisation.
Since court records often contain highly sensitive personal information, judicial AI deployments frequently have to satisfy both the AI Act and GDPR.
The result is a regulatory framework that neither prohibits nor fully embraces AI-driven justice. Instead, the EU seeks to ensure that AI remains a tool used by judges rather than a substitute for them, with extensive safeguards intended to preserve accountability, transparency and public trust in the judicial process.
America’s guidance model
The United States, on the other hand, uses a much looser guidance model, propagating guidelines without expressly restricting usage.
The country has a chequered history with the use of software in courts.
For years, many states in the U.S. have used simpler algorithmic tools to assign offenders as ‘high-risk’ or ‘low-risk’, and these classifications have then influenced bail decisions and even sentencing decisions.
These programmes, however, became controversial with multiple investigations by the media alleging racially biased outcomes.
Unlike the EU, the U.S. has no single nationwide framework governing the use of AI in courts. Instead, guidance has emerged through a patchwork of judicial conferences, court rules, bar associations and individual judges’ orders.
The focus has generally been less on regulating AI systems themselves and more on regulating how lawyers and litigants use them.
In 2023, for instance, several lawyers were sanctioned after citing fictitious cases generated by ChatGPT in court filings. Many courts began requiring lawyers to verify AI-assisted research and sign declarations affirming that all citations have been independently checked.
In the absence of a truly national policy, some states have begun adopting guidelines. The state of Delaware, a hub for corporate litigation, adopted policies stating that the use of AI for research is permitted, but court officers are “responsible for the accuracy” of everything they generate.
The state of New York’s policies have emphasised education, providing for “mandatory initial and continuing training” for judicial officers, which is meant to hammer home the principle that “judges remain responsible for the content of their decisions”.
These guidelines, however, remain largely advisory, and enforcement mechanisms vary. The US Supreme Court has not yet issued a landmark ruling defining how generative AI may be used in court proceedings or setting nationwide standards for responsibility and disclosure, leaving regulation fragmented across federal and state courts.
China’s smart courts
China’s policies mark a divergence from those of the West and represent the most ambitious attempt to integrate AI into a judicial system through the use of “smart courts”.
As far back as 2017, a Chinese government white paper noted its aim to “promote AI applications for applications including evidence collection, case analysis, and legal document reading and analysis”.
Since then, courts have been encouraged, and even required, to develop AI tools.
The effects of this policy are apparent both nationally and locally.
At the local level, courts have used AI, trained on case databases, to create ‘similar-case’ systems that can predict based on precedent. Shanghai’s system, for instance, “learns from past cases to give sentencing predictions and recommendations.” according to researchers at Oxford University.
On a larger scale, the Supreme People’s Court, the country’s highest judicial body, has developed “a national-level legal AI infrastructure built on massive, authoritative and high-quality judicial data”, according to a press release.
This database, which holds hundreds of millions of cases, is meant for the use of regional courts to train their own AI systems like Shanghai’s.
China’s goals appear to be two-fold. On the one hand, the systems are part of a larger push towards judicial consistency across the vast country. On the other, the program is a response to the mounting backlog in China’s courts.
While the country is becoming more litigious, the number of judges has decreased. As part of its reforms in 2013-14, China reduced the number of judges substantially from 200,000 to around 120,000.
Its AI policies are now geared towards streamlining and accelerating the disposal of cases with the extensive use of AI.
Where does India fit in?
India’s draft framework sits somewhere between Europe’s governance-heavy model and the US’s guidance-based approach, while remaining far more cautious than China’s smart-courts programme.
Like the EU, the draft emphasises human oversight, accountability and transparency. Like the US, it places ultimate responsibility on judges, lawyers and litigants rather than on the AI systems themselves.
But unlike China, it does not envision AI playing a substantive role in adjudication. AI is treated as an assistive tool for functions such as research, translation and case management, not judicial decision-making.
What is striking, however, is that none of these frameworks seek to completely exclude AI from the courtroom. While jurisdictions differ on how tightly it should be regulated, they increasingly appear to agree on one point: AI’s presence in the justice system is inevitable, and the challenge is ensuring that human judgment remains in control.
Sahaj Sankaran is a TPSJ alum, currently interning with ThePrint.
(Edited by Sugita Katyal)
Also read: Law professors prefer AI over colleagues’ answers 75% of the time, says Stanford study