Drafting Indian-Law Contracts With AI: Where Global Tools Get It Wrong

Ask a general-purpose AI tool to draft an Indian employment offer letter, and there is a good chance it hands you a clause restraining the employee from joining a competitor for a year after they leave. That clause is void the day it is signed. Ask it to draft an IP assignment for a contractor, and it will likely say nothing about how long the assignment lasts, which under Indian copyright law does not mean forever, it means five years. These are not edge cases. They are the default output of a model trained mostly on US and UK contracts, applied to a legal system that runs on different rules, because the public internet's English-language contract text is overwhelmingly American. (Adira, which publishes this guide, builds contract drafting and review software, including tools grounded in Indian statute and your own templates, so we have a commercial interest in you knowing this. Everything below is written to be useful whether or not you ever use our product.)

This guide walks through six specific ways this goes wrong in an Indian-law contract, with the actual statute quoted for each, so you can check any AI draft yourself before it goes near a signature. We cover the general reasons AI drafts hallucinate and get commercial terms wrong, including case citations Indian courts have now sanctioned lawyers for using, in Can AI Draft a Contract? What It Does Well and Where It Fails. This page is narrower: the India-law traps, one by one, statute in hand.

Trap 1: A post-employment non-compete that is void the day it is signed

What the AI hands you: language restraining an employee or contractor from working for a competitor for six to eighteen months after they leave, dressed up as "reasonable" because it names a time limit and a geography, the way a US non-compete does.

Why it fails in India: Section 27 of the Indian Contract Act, 1872 says:

"Every agreement by which any one is restrained from exercising a lawful profession, trade or business of any kind, is to that extent void."

Read the section on Indian Kanoon. India runs no "reasonableness" test for post-employment restraints the way US courts do. The restraint is void outright, with a narrow exception for the sale of business goodwill. The Delhi High Court applied this again in 2025, refusing to enforce a post-termination restrictive covenant against a departing employee. We cover that case and how to spot a disguised non-compete hiding inside a "non-solicitation" clause in full in Are Non-Compete Clauses Enforceable in India?

The test: search the draft for any restraint that operates "following termination," "after the Agreement ends," or "for [X] months post-employment." If it does, and it is not limited to confidentiality or a business-sale scenario, treat it as void under Section 27, whatever the time period says.

Trap 2: An IP assignment silent on period and territory, which does not mean forever

What the AI hands you: "the Contractor hereby assigns to the Company all right, title and interest in the Work Product," with no line stating how long the assignment lasts or where it applies, the US "work made for hire" pattern where silence usually means permanent, worldwide ownership.

Why it fails in India: Section 19(5) of the Copyright Act, 1957 says:

"If the period of assignment is not stated, it shall be deemed to be five years from the date of assignment."

And Section 19(6) says:

"If the territorial extent of assignment of the rights is not specified, it shall be presumed to extend within India."

Read both on Indian Kanoon. This is not theoretical. In Pine Labs Pvt. Ltd. v Gemalto Terminals India Pvt. Ltd. (Delhi High Court, 2011), the assignment clause used the word "assigns" but stated no period or territory, and the court held Sections 19(5) and 19(6) were "inevitably triggered" by that silence. We cover the full mechanics, including the Section 17 employer default, in IP Assignment Clauses in India: The 5-Year Default Most People Miss

The test: search the assignment clause for "in perpetuity" and "throughout the world" or an equivalent. If neither phrase appears, the default clock has already started.

Trap 3: "At-will" employment does not exist here

US templates routinely include an at-will clause, letting either party end employment at any time, for any reason or none, with no notice. There is no Indian equivalent. Every Indian employment relationship runs on a stated or statutory notice period, and for employees who qualify as "workmen" under the Industrial Disputes Act, 1947, Section 25F adds a further layer: no workman with at least one year of continuous service can be retrenched without one month's written notice, or pay in lieu, plus retrenchment compensation. State Shops and Establishments Acts add their own notice rules for other staff. An AI that pastes in an at-will clause is not describing a lighter version of Indian practice, it is describing a concept Indian labour law has no room for. The safer default is an explicit notice period, commonly 30 to 90 days, stated in the contract itself.

The test: search the offer letter or employment contract for the word "at-will." If it appears, and the contract does not separately state a notice period, that clause is describing US law, not the law that actually governs the relationship.

Trap 4: No awareness that the document may need stamping or registration

A US or UK drafting tool has no reason to know Indian instruments can require stamp duty, and sometimes registration, to be usable at all. Section 35 of the Indian Stamp Act, 1899 says:

"No instrument chargeable with duty shall be admitted in evidence for any purpose by any person having by law or consent of parties authority to receive evidence, or shall be acted upon, registered or authenticated by any such person or by any public officer, unless such instrument is duly stamped."

Read it on Indian Kanoon. An unstamped agreement is not automatically void, but a court or arbitrator cannot act on it or admit it in evidence until duty, and often a penalty, is paid, which can stall a dispute at the worst possible moment. Separately, Section 17(1)(d) of the Registration Act, 1908 requires compulsory registration of "leases of immovable property from year to year, or for any term exceeding one year, or reserving a yearly rent" (read it here). Rates and thresholds vary by state and instrument, so no AI tool, and no generic guide including this one, should hand you a specific rate with confidence. It should just flag that the question exists.

The test: if the document is a lease, a leave-and-licence agreement, a conveyance, or anything involving immovable property or a term over a year, check whether it needs stamping and registration in the relevant state before treating it as final, regardless of what the AI draft says or does not say about it.

Trap 5: US-style "penalty vs liquidated damages" framing that India does not use

What the AI hands you: careful language insisting a damages clause is a "genuine pre-estimate of loss," with a note that an uncapped or round-number penalty risks being struck down, the English and US two-track rule where a "penalty" is void but true liquidated damages are enforced in full.

Why it fails in India: Section 74 of the Indian Contract Act, 1872 collapses that distinction. It says the party complaining of breach:

"is entitled, whether or not actual damage or loss is proved to have been caused thereby, to receive from the party who has broken the contract reasonable compensation not exceeding the amount so named or, as the case may be, the penalty stipulated for."

Read it on Indian Kanoon. Whether the contract calls the sum "liquidated damages" or "penalty," an Indian court treats both the same way, a ceiling on reasonable compensation, not an entitlement to the full amount. The Supreme Court applied this in Kailash Nath Associates v DDA ((2015) 4 SCC 136), where a forfeiture clause calling itself automatic on breach still failed because no actual loss was shown. We cover that case and ONGC v Saw Pipes Ltd ((2003) 5 SCC 705) in full in Liquidated Damages vs Penalty in India

The test: if the AI draft spends effort distinguishing "penalty" from "liquidated damages" as if one survives and the other does not, that is the English or US rule leaking in. Under Indian law, both labels get the same treatment.

Trap 6: Governing law and jurisdiction defaults that quietly fail

Two patterns show up often. Governing law defaults to Delaware, California, or New York, the most common choice-of-law clause in the training data, with no connection to either party. And jurisdiction clauses name multiple courts "for convenience" or use vague language, when Indian law is fairly strict about what makes an exclusive jurisdiction clause actually exclusive. Section 28 of the Indian Contract Act, 1872 voids any agreement that restricts a party "absolutely" from enforcing rights through the ordinary courts. But where two or more courts would otherwise have jurisdiction under the Code of Civil Procedure, the Supreme Court held in A.B.C. Laminart Pvt. Ltd. v A.P. Agencies, Salem ((1989) 2 SCC 163) that parties can validly restrict disputes to one of those courts, without violating Section 28, as long as they do not hand jurisdiction to a court that never had it. No magic word like "exclusive" was required, just clear intent, which a clause naming an unconnected court, or leaving the choice ambiguous between several, usually lacks.

The test: check whether the named governing law and courts actually connect to where the parties are based or the contract is performed. A US state with no party presence, or an ambiguous "courts of India shall have jurisdiction" with no city named, is a default worth rewriting.

Red flags in an AI-drafted Indian contract

Bad clause, rewritten for India

What an AI often produces for an Indian employment offer letter, unprompted:

"Employment under this offer letter is at-will and may be terminated by either party at any time, with or without cause, and with or without notice. For a period of twelve (12) months following termination, the Employee shall not, directly or indirectly, provide services to any business that competes with the Company."

Two failures in two sentences: "at-will" describes a doctrine India does not have, and the twelve-month restraint is void under Section 27 regardless of how it is worded.

A version corrected for India:

"Either party may terminate this employment by giving 60 days' written notice, or payment in lieu of notice, subject to any additional notice or compensation required under applicable law for the Employee's category of employment. The Employee's obligations of confidentiality under Clause 9 survive termination for 24 months. This Agreement does not restrict the Employee from taking up employment with any other organisation, including a competitor, after termination."

What changed: notice replaces "at-will," the post-termination restraint is removed rather than left to be struck down later, and the protection the company actually needs, confidentiality, sits in its own surviving clause instead of being smuggled inside an unenforceable non-compete.

Why grounding the AI in Indian law and your own corpus matters

Every trap above has the same root cause: a generic model, given a generic prompt, reproduces the statistically dominant pattern in its training data, a US or UK pattern. Fixing this needs two layers: grounding the AI in Indian statute itself, so it starts from Section 27, Section 19, and Section 74 instead of an internet-wide average, and grounding it in your own corpus, past contracts and negotiating positions, so it drafts in your house style on top of a legally correct base. We cover how corpus grounding works, and its own honest limits, it fixes genericness, it does not by itself verify enforceability, in What Is a Company Legal Persona?

Adira's drafting tools start from Indian statute for the areas above, rather than a US-trained default, and layer your own templates on top through that same persona mechanism. That reduces how often these six traps show up in a first draft; it does not remove the need for a human to check enforceability on anything with real stakes. To sanity-check a single clause first, without any of this, mark it up for free in Weave, Adira's free browser-based contract tool, and check it against the red flags above.

FAQ

Is every AI-drafted Indian contract wrong? No. Boilerplate like notices, severability, and counterparts clauses is close to interchangeable across jurisdictions and usually fine. The failures cluster around the six areas above, where Indian law diverges sharply from the US patterns most training data reflects.

Can I just tell the AI "use Indian law" in my prompt and fix this? It helps, but it is not reliable on its own for a tool with no Indian legal grounding, since the model still generates from the same training data with an instruction on top that it may not apply consistently across a long document. Checking the draft against the statutes above is safer than trusting the instruction alone.

Does a non-compete become enforceable if I get the employee's separate written consent to it after they resign? No. Section 27 voids the restraint itself, regardless of consent given at signing or afterward. The fix is not including a post-termination restraint at all, and relying on confidentiality, and where genuinely justified, garden leave during employment instead.

If my IP assignment clause already says "in perpetuity, worldwide," do I still need to worry about Section 19? No, for that clause. Section 19(5) and 19(6) only apply where the period or territory is unstated. The trap is specifically silence, which is what most AI-generated "work for hire" style clauses default to.

Will Adira's tools always catch every one of these traps automatically? No tool should be trusted to catch everything automatically, and Adira does not claim that. Grounding in Indian statute and your own corpus reduces how often these patterns appear in a first draft; it does not replace a lawyer's review before anything with real stakes.

This guide gets you to spotting the patterns where a US or UK-trained AI tool produces a wrong Indian contract. It does not tell you whether a specific clause in your contract is enforceable, whether stamping or registration applies in your state, or whether a damages figure would survive challenge. For that, especially on anything above a low-stakes document, have an Indian lawyer review the draft before you sign it. This is not legal advice.