India: what courts have said about AI in legal work
By the ChamberOne team 5 minute read
Indian courts are still working out their position on artificial intelligence, and the position is not uniform. Some High Courts have published policies for their own judges and staff. The Supreme Court and other bodies have commented in the course of proceedings and in public papers. This article sets out what can be said with confidence from official and well-reported sources, is plain about what it cannot say, and suggests how advocates and firm managers should behave while the picture develops. It is general information, not legal advice.
Key points
- Judicial policy on AI in India is developing and differs between courts, so check the current primary source.
- The Kerala High Court policy for the district judiciary draws a clear line against using AI to reach findings, orders or judgments.
- Advocates remain responsible for every citation they file, whatever tool suggested it.
- Firms should set a written policy, train juniors and keep a note of how authorities were verified.
A moving picture, so read the primary source
The first thing to say is that anything written about court policy on AI dates quickly. A circular issued last year may have been replaced, and a remark in one bench's order may not reflect the view of another. Before you rely on any statement in this article, check the current position on the website of the relevant High Court or the Supreme Court, and look for notices from the Bar Council of India and your state bar council.
Another point is that policies for judges are not the same as rules for advocates. A court telling its own judges what tools they may use does not by itself set the duties of lawyers who appear before it. It does, however, show what the court is worried about, and sensible advocates read it that way.
The Kerala High Court policy
The best-documented example is the policy of the Kerala High Court on the use of artificial intelligence tools in the district judiciary, issued in July 2025 and reported as the first of its kind in India. It applies to members of the district judiciary in the State and to the employees who assist them.
As reported, its central instruction is that AI tools must not be used to arrive at findings, reliefs, orders or judgments, and must not stand in for decision-making or legal reasoning. The policy also requires officers and staff to take part in training sessions. Reports of the policy mention a preference for tools approved by the courts and caution about public cloud-based services, but you should read the text of the policy itself for the exact wording and conditions.
What the caution on decision-making tells you
The line the policy draws is useful to advocates, because it separates assistance from judgment. Administrative help, such as organising material or translating for convenience, is treated differently from the core judicial function of deciding. The reasoning follows a theme found in other countries' judicial guidance: the person who decides must be a person, and remains answerable for the result.
For an advocate, the implication is straightforward. A court expects that submissions are the work of the lawyer who signs them, and that the lawyer has verified what they say. Using a tool for a first draft is one thing. Presenting its output as checked when it is not is another.
- Organising material
- Translating for convenience
- A first draft, verified by the lawyer who signs
- Reaching findings, reliefs, orders or judgments
- The person who decides must be a person
- They remain answerable for the result
Fabricated authorities and professional responsibility
Across many jurisdictions, courts have been confronted with submissions that cite judgments that do not exist, produced by AI tools. Indian courts and commentators have taken notice of this problem too, and the discussion has turned to the professional responsibility of the advocate who files such material. Reports of the exact orders, committees and papers vary in date and detail, so look at the orders themselves.
Whatever the final shape of any rule, the practical lesson holds already. Every citation in a document you file should be one you or a colleague has found in a recognised law report or database, read and confirmed to support the proposition. The tool that suggested it is not a source.
Treat any fresh development, such as a new circular or a committee report, as a reason to review your own checking steps, not as a reason to wait.
What advocates should do now
A cautious working practice, which you can adapt, looks like this.
- 1
Check each court you appear in for notices, circulars or practice directions on AI, and keep copies on the file.
- 2
Check your bar council's guidance, and the guidance from your regulator or bar association, before using an AI tool on client work.
- 3
Verify every authority and every quotation against a recognised report or the official text before it goes into a filing.
- 4
Keep a short note on the file saying how citations were verified.
- 5
Tell the client if you intend to use AI on their matter and note any instruction they give you.
- 6
Do not put confidential client information into a public AI service.
- Check each court for notices and circulars on AI
- Verify every authority and quotation before filing
- Keep a note of how citations were verified
- Tell the client if you intend to use AI
- Put confidential client information into a public AI service
- Present output as checked when it is not
What firm managers should do
Managers should set a written policy that names the approved tools, lists what may never be entered, requires verification of citations and says who answers questions. Train juniors in particular, since they often draft first and are least likely to challenge a confident answer. Keep an eye on language and translation: Indian practice often involves several languages, and an AI translation of a document or order can drift from the original in ways that matter. Check translations with a person who reads both languages.
Think also about data. Client files may include personal information, and the law on personal data in India has been developing. Take advice on how it applies to outside AI services before you send client material to one.
Keeping up
Assign one person in the firm to follow developments: new High Court policies, Supreme Court observations, bar council notices and any legislation. A short monthly note to all fee earners is usually enough. Where a court issues a policy, read the whole text rather than a news summary, because the conditions and exceptions are where the practical rules sit.
Until the position settles, the sound approach is the one courts themselves appear to favour: human decision-making, human verification and care with confidential information.
Consider keeping a simple register of court notices, with the court, the date, a one-line summary and where the full text is stored, so a new joiner can see at a glance what the firm has seen and acted on.
Sources
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
India's new criminal codes: BNS, BNSS and BSA in practice
What India's three new criminal codes mean in practice: the transition rule, key substantive changes, procedural shifts and electronic evidence reforms.
6 minute readGenerative AI guidance for lawyers: how six markets compare
How to find what regulators, courts and bar bodies have said in six markets, and the common themes across them.
5 minute readEU AI Act: what it means for firms outside the EU with EU clients
How a firm outside the EU should think about the EU AI Act when it works for European clients.
5 minute read