What Jurisdiction-Aware Legal AI Means for Indian Practice

Legal AI that is not jurisdiction-aware is not merely less useful. It is wrong in a way that reads as right. Law is territorial: the same sentence in a contract can be routinely enforced in one country and void by statute in another, and a model whose training material is weighted towards American and English legal text can answer an Indian question fluently, confidently, and in foreign law. This piece sets out what jurisdiction-awareness has to mean concretely, and why the absence of it is a professional hazard rather than a quality-of-life issue.

The same clause, two opposite outcomes

Take a handful of clauses that appear in nearly every commercial agreement, where the Indian position diverges sharply from what a US or UK-weighted system is likely to produce.

• Post-termination non-compete. Section 27 of the Indian Contract Act 1872 makes an agreement by which a person is restrained from exercising a lawful profession, trade or business void to that extent, subject to a statutory exception concerning the sale of goodwill. Indian courts have generally declined to enforce post-employment restraints on this basis, while treating restraints operating during the subsistence of employment differently. A reasonableness test imported from another legal system does not rescue the clause.
• Liquidated damages. Section 74 of the Indian Contract Act 1872 entitles the aggrieved party to reasonable compensation not exceeding the amount named in the contract, whether or not actual damage is proved. That statutory text is qualified by case law: Indian courts have held that compensation under Section 74 requires loss or damage occasioned by the breach, with the relaxation on proof operating principally where loss is difficult to quantify, and that the named sum operates as a ceiling rather than an automatic entitlement. The Indian statutory scheme does not track the English penalty versus genuine pre-estimate distinction in the same way, and drafting that assumes it will can produce a clause that does not operate as intended.
• Indemnity. Section 124 of the Indian Contract Act 1872 defines a contract of indemnity in narrower statutory terms than the expansive commercial usage common in cross-border drafting, and Section 125 sets out the rights of the indemnity-holder when sued. Both affect how broadly a transplanted indemnity can be relied on.
• Electronic execution. Section 1(4) read with the First Schedule to the Information Technology Act 2000 excludes specified instruments from the Act's application, a list that has historically included negotiable instruments other than cheques, powers of attorney, trusts, wills and contracts for the sale or conveyance of immovable property. The Schedule is amendable by notification, so confirm the current position against the notification in force. A workflow designed abroad on the assumption that everything can be signed electronically can generate an instrument that fails here.
• Stamping. Stamping in India is substantially a State subject. Some States apply the Indian Stamp Act 1899 as amended in its application to that State; others have their own legislation, such as the Maharashtra Stamp Act 1958 or the Karnataka Stamp Act 1957. Insufficient stamping carries admissibility consequences, flowing from Section 35 of the Indian Stamp Act 1899 or the corresponding provision of the applicable State Act, and rates, articles and adjudication procedure differ between States. Identify which State's stamp law applies before assessing duty or admissibility, and check the rate against the schedule currently in force in that State. There is no directly comparable step in ordinary US contracting practice, so a system trained largely on foreign material may not raise it at all.

None of these are exotic edge cases. They are the first two pages of an ordinary services agreement, and each of them is a place where a confident foreign answer produces an unenforceable document.

Six things a jurisdiction-aware system has to track

Jurisdiction-awareness is not a single flag on a query. It is a set of coordinates the system must hold at all times, because changing any one of them can change the answer.

Applicable law

The governing law clause is the starting point, not the answer. Mandatory Indian law can apply regardless of what the parties chose: stamp duty on instruments executed or received in India under the applicable central or State stamp legislation, exchange control under the Foreign Exchange Management Act 1999, withholding obligations under the governing income-tax legislation, statutory employment protections, and public policy limits on enforcement of foreign judgments and awards. A capable system distinguishes what law the parties selected from what law applies anyway.

Forum

Which court or tribunal will decide, and what binds it. Article 141 of the Constitution makes the law declared by the Supreme Court binding on all courts in India, but below that the picture is territorial: a High Court's ruling binds courts subordinate to it and is persuasive elsewhere. The NCLT, ITAT, CESTAT and CCI each sit in their own appellate structures. A system that returns relevant-looking cases ranked by textual similarity, without ranking by binding force on the specific forum, has produced a reading list rather than research.

Regulator

Compliance obligations attach to named regulators, not to abstractions. In India, a single transaction can engage SEBI through listed-entity disclosure obligations under the SEBI (Listing Obligations and Disclosure Requirements) Regulations 2015, the RBI through FEMA 1999 and foreign investment reporting, the MCA and the Registrar of Companies through filings and registrations under the Companies Act 2013 including charge registration under Section 77 within the prescribed period, and the GST authorities through place-of-supply determination and export conditions under the IGST Act 2017. None of these map onto the SEC, the FCA or HMRC, and a system reasoning by analogy to those bodies can miss the obligation entirely.

Effective date

Law has a version history. The question is rarely what a section says today and almost always what it said on the date that matters: the date of the contract, of the transaction, of the assessment year, of the alleged offence. A system that only knows the currently in-force text will answer historical questions wrongly and without hesitation, which is worse than not answering them.

Document type

What a document is determines what attaches to it. An employment agreement, a share subscription agreement, a related-party transaction, a lease of immovable property and a standalone arbitration agreement raise different mandatory requirements around stamping, registration, board and shareholder approvals, and regulatory filing. Clause-level analysis without document-type awareness reliably misses the requirement that sits outside the four corners of the clause being reviewed.

Entity location

Place of incorporation, place of business, counterparty location, place of performance and location of data all move the answer. An Indian entity paying a non-resident engages withholding obligations, and questions of treaty relief, under the governing income-tax legislation. Because the Income-tax Act 2025 has replaced the Income-tax Act 1961, section numbers drawn from the older Act should not be treated as current: identify the corresponding provision in the Act now in force and verify it against the bare Act before relying on it. An Indian subsidiary of a foreign group will often be operating under group policies drafted for a different regulatory regime, which is precisely where silent mismatches accumulate.

Why a model weighted towards US and UK text is a hazard here

Where this goes wrong, the failure mode is typically not that the model declines to answer. It is that it substitutes silently, filling an Indian question with the nearest foreign concept it knows well.

• Default vocabulary from another system: at-will employment, Delaware, the UCC, best-efforts standards, doctrines with no Indian statutory analogue.
• Confidently produced section numbers from the wrong statute, or Indian-sounding provisions that do not exist in the form claimed.
• Analysis that omits steps with no foreign counterpart: stamping, registration, exchange control, and where applicable prior regulatory approval.
• Foreign decisions offered as authority for Indian propositions. English judgments can be persuasive, but they do not bind an Indian court, and where an Indian statute governs the field the statute prevails over imported common law.
• Terminology collisions. Consideration, indemnity, specific performance, injunction and discovery all carry different statutory content in India from their apparent equivalents elsewhere.

This is a hazard rather than an inconvenience because fluent foreign law is far harder to catch than obvious nonsense. A junior can spot a fabricated case name. It takes considerably more experience to notice that an otherwise sensible memo quietly assumed a doctrine that Indian statute has displaced, and that assumption then propagates into the drafting.

A model that answers an Indian question in American law does not sound uncertain. That is precisely the problem.

Time is a jurisdiction dimension too

India has passed through a period of rapid statutory change, which makes effective-date awareness more than a refinement. The Bharatiya Nyaya Sanhita 2023, the Bharatiya Nagarik Suraksha Sanhita 2023 and the Bharatiya Sakshya Adhiniyam 2023 replaced the Indian Penal Code 1860, the Code of Criminal Procedure 1973 and the Indian Evidence Act 1872 with effect from 1 July 2024, which means section numbering in decades of case law no longer maps directly onto the provisions now in force. The Companies Act 2013 replaced the Companies Act 1956. The Income-tax Act 2025 replaced the Income-tax Act 1961, so income-tax section numbers in older judgments, opinions and precedent files cannot be assumed to correspond to the provisions now in force; identify the current provision in the bare Act rather than carrying the old number forward. GST replaced a layered indirect tax structure in 2017, and rate and place-of-supply positions continue to move by notification, so verify any rate or threshold against the notification currently in force.

On data protection, the Digital Personal Data Protection Act 2023 received Presidential assent in August 2023, and the rules made under the Act have since been notified. The framework does not switch on all at once: different provisions and obligations take effect at different points, and that sequencing sits in the commencement and notification provisions rather than being apparent from the Act alone. Confirm which provisions are in force, and the date by which each obligation bites for your organisation, against the Gazette notification and the material published by the Ministry of Electronics and Information Technology before committing to any compliance date. The broader point holds independently of where that sequencing has reached: a cross-border privacy programme built on GDPR assumptions does not transfer to the Indian framework unexamined, because the two regimes take different structural approaches to matters including cross-border transfer.

What good looks like in practice

Jurisdiction-awareness is testable. These are the properties worth insisting on when evaluating any legal AI tool for Indian work, whoever built it.

• Every substantive answer names the jurisdiction, the statute and section relied on, and the version or date of the law applied.
• Authority is ranked by binding force on the specific forum, not by textual similarity to the query.
• Foreign material, where surfaced, is labelled as foreign and treated as persuasive at best.
• The system knows which regulator a given obligation belongs to, and can produce the filing calendar that follows from it.
• Effective-date filtering is a first-class control rather than something the user has to remember to apply.
• Gaps in coverage are reported as gaps, not filled with the closest available answer.

LexVio is built on Indian primary sources. Court research covers the Supreme Court, High Courts, NCLT, ITAT, CCI and CESTAT, with filtering by jurisdiction, search by clause type and citation graph visualisation. Compliance monitoring covers SEBI, RBI, MCA and ROC, and GST, with an automated filing calendar, 14-day pre-deadline alerts and a regulatory change feed. Tax AI covers income tax, TDS and GST computation, IFRS versus IndAS comparison, and cross-border tax impact. For portfolio-level work, Nexus supports search across a document set, clause benchmarking and drift alerts, while Vio works on a single document at a time. Teams with data control or deployment requirements can use SAML SSO, SCIM provisioning and an on-prem deployment option.

None of this shifts the professional judgment. The system compresses the mechanical work of locating candidate provisions, filtering by forum and tracking deadlines; the decision about what to advise, what to sign and what to file stays with the lawyer, the CA or the CS. That division is not a limitation to be engineered away. It is what makes the tooling usable in regulated practice at all.

There is a simple test for any legal AI product being pitched for Indian work. Ask it a question where the Indian answer differs from the American one, such as the enforceability of a post-termination non-compete, or whether a particular instrument can be executed electronically. Then see whether it notices the difference on its own, or only after you tell it which country you are in.