Ban on AI neither practicable nor wise, says CJI Surya Kant
Surya Kant argued that a blanket ban on generative AI in legal work and education would be neither “practicable nor wise.” These tools can summarise judgments, suggest legal arguments and produce polished answers quickly. He said AI is already changing legal education, research and professional work, so institutions must address its use rather than pretend it can be stopped. The CJI said technological help must not replace independent reasoning, verification or accountability. For students, the key issue is whether submitted work still shows their own understanding of the law. He pointed to Singapore’s Supreme Court as an example: it permits court users to use generative AI but makes them responsible for everything they file. Kant’s recommended approach is to establish clear rules and teach responsible use. He also called on universities, academic supervisors and publishers to create guidance for AI in research. This would allow technology to assist legal work while keeping people answerable for its accuracy and integrity.
What did Chief Justice Surya Kant say about banning generative AI tools in legal work and education?
Surya Kant argued that a blanket ban on generative AI in legal work and education would be neither “practicable nor wise.” These tools can summarise judgments, suggest legal arguments and produce polished answers quickly. He said AI is already changing legal education, research and professional work, so institutions must address its use rather than pretend it can be stopped.
The CJI said technological help must not replace independent reasoning, verification or accountability. For students, the key issue is whether submitted work still shows their own understanding of the law. He pointed to Singapore’s Supreme Court as an example: it permits court users to use generative AI but makes them responsible for everything they file.
Kant’s recommended approach is to establish clear rules and teach responsible use. He also called on universities, academic supervisors and publishers to create guidance for AI in research. This would allow technology to assist legal work while keeping people answerable for its accuracy and integrity.
What is generative AI, and how many different kinds of legal tasks can it help with?
Generative AI refers here to tools that produce responses or written material, rather than only retrieving a fixed document. The article focuses on how such tools can be used in legal settings. It does not provide a technical account of how the systems work; it describes what they can do for users.
It gives three examples of legal tasks: summarising judgments, suggesting legal arguments and producing polished answers within minutes. These outputs can help students and professionals handle legal material, but a convincing response is not necessarily a correct one. The tools may also miss important qualifications or authorities.
That distinction matters because legal work depends on accurate sources and careful reasoning. Kant said technological assistance should support a student’s search, not replace independent thought or accountability. The article’s approach is not to reject these tools, but to pair their use with clear rules, verification and responsibility for the final work.
Why did the CJI argue that a ban could push AI use underground, and what alternative did he recommend?
Kant’s concern was that prohibiting AI would not necessarily stop students or legal professionals from using it. Instead, he said, use could move underground. If that happened, institutions would have less opportunity to guide how people use the tools or examine the work they produce.
His alternative is to set clear rules and teach responsible use. The article gives Singapore’s Supreme Court as an example of a different approach: court users may use generative AI, but they remain fully responsible for everything they file. This places accountability on the person submitting the work, not on the tool.
Kant said law schools should reconsider assessments and ask whether students’ work reflects their own reasoning, understanding and appreciation of law. He also urged universities, supervisors and publishers to develop guidance for research. The aim is to let technology assist legal work while preserving human judgment, verification and responsibility.
What can go wrong when lawyers or students rely on AI for legal research?
The main risk is that AI-generated legal research can sound authoritative without being reliable. Kant warned that a fluent and convincing answer is not necessarily accurate. That matters in law, where a false authority or an incomplete account of a judgment can distort an argument or mislead readers.
The article identifies several specific problems. AI may fabricate judicial authorities, omit crucial qualifications when summarising judgments, or produce incomplete comparisons by missing decisions from parts of the Commonwealth that are poorly represented in digital databases. A polished response can therefore conceal gaps or errors. Researchers are expected to check each authority against its primary source.
Kant also said researchers must disclose material help received from AI and remain answerable for what they publish. The tools cannot be credited as authors because they cannot take responsibility for the work. His warning is not that every AI output is wrong, but that users must verify it rather than rely on its confident style.
Under the approach described for Singapore's Supreme Court, who is responsible for material filed with help from AI?
Under the approach Kant described, using AI does not transfer responsibility from the person submitting legal material to the technology. The Supreme Court of Singapore permits court users to employ generative AI, but those users remain fully responsible for everything they file.
That rule addresses a central risk of AI-assisted legal work: a tool may produce polished text that contains fabricated authorities, omissions or other errors. The human user must therefore check the work rather than treat the AI system as an accountable contributor. Kant said an AI system cannot be credited as an author precisely because it cannot be held responsible.
The Singapore example supports Kant’s broader argument against a blanket prohibition. Rather than banning these tools, institutions can set rules that make clear who must stand behind submitted work. In the article, that responsibility rests with the court user. Kant also urged legal educators and researchers to keep independent reasoning and verification central to AI-assisted work.
What should researchers check or protect before using AI for legal work?
Kant said researchers must verify every authority produced or cited with AI against its primary source. This matters because AI can fabricate judicial authorities, leave out qualifications in judgment summaries or overlook relevant jurisprudence. A fluent response alone is not proof that its legal claims are accurate.
Researchers should also disclose material assistance received from AI tools and remain answerable for what they publish. Before entering information into an external platform, they must consider whether it is confidential. The CJI specifically cautioned against uploading unpublished manuscripts, research data or client information without due regard to consent, privacy and security.
He called for universities, academic supervisors and publishers to create clear guidance on AI use in research. Such rules can help people distinguish acceptable assistance from risks to accuracy or confidentiality. Kant’s message is that responsible use involves both checking what AI produces and taking care over what information users provide to it.
Why should legal education assess students' own reasoning and ability to apply law, rather than just polished written answers?
Kant said law schools must ask whether students’ work reflects their own reasoning, understanding and appreciation of law. That matters because generative AI can produce polished answers quickly, but a smooth response does not prove that a student has understood the legal issues or reached the reasoning independently.
He also argued that legal education should prepare students to apply legal principles to real disputes, rather than depend excessively on theory. Practical learning, he said, should reach core subjects such as evidence, criminal procedure, family law and constitutional law, not remain limited to separate clinical courses. He gave examples of drafting and negotiating an agreement in contract law and tracing a denied entitlement to a legal remedy in constitutional law.
The article does not prescribe a specific assessment system. Kant’s direction is to rethink assessment and teach responsible AI use, so technology can assist learning without replacing students’ own judgment. This would help prepare them for a profession already adopting these tools.
What does an AI “hallucination” mean in the context of legal research?
In this context, an AI “hallucination” means an output that appears credible but contains invented or inaccurate legal information. The article does not use the term itself, but it describes the problem: generative AI can produce fabricated judicial authorities. Its fluent style may make the error harder to spot.
The risks go beyond invented cases. Kant said AI may omit crucial qualifications when summarising judgments or produce incomplete comparative analyses by overlooking jurisprudence from parts of the Commonwealth that digital databases represent poorly. These gaps can make research misleading even when the response reads clearly and confidently.
Kant’s safeguard is direct verification: researchers should check every authority against its primary source, disclose material AI assistance and remain responsible for what they publish. The practical lesson is not to assume that a plausible legal answer is accurate. People using AI must examine its claims before relying on them in research or professional work.
This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.
Read more in the JupiteX app
Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.
Or read more news on the web