Clauseium vs ChatGPT — Is ChatGPT Safe for Indian Contract Review?

TL;DR

ChatGPT is a general-purpose AI assistant. It is not safe for production Indian contract review for three reasons:

1. Hallucinated citations — ChatGPT invents Section numbers, cites repealed statutes, and fabricates case names. Research shows hallucination rates of 25-60% on legal queries. There is no verification step; the prose is fluent so errors are hard to spot.
2. No Indian-law grounding — ChatGPT is trained on a global corpus that under-represents Indian legal materials. It treats the Indian Contract Act, 1872 with the same depth as a state statute from Wyoming.
3. Confidentiality and DPDP risks — Standard ChatGPT may train on inputs and stores history in US data centres. For an Indian company processing personal data, this raises Section 8(5) processor-contract issues and Section 16 cross-border-transfer concerns.

ChatGPT is excellent for brainstorming, drafting first-cut prose, and summarising long documents. It is not a contract review tool. Clauseium is purpose-built for the contract-review workload, with citation verification, Indian-law grounding, and DPDP-compliant data handling built in.

Why this comparison comes up so often

Indian in-house counsel ask "can I just use ChatGPT?" because ChatGPT is cheap (or free) and ubiquitous. The question is fair. The answer is: depends what you mean by "use."

For research, brainstorming, drafting boilerplate, summarising long documents — ChatGPT is fine. For client-facing contract review where the output gets relied upon — ChatGPT is not safe. The difference matters because the failure modes are silent. ChatGPT will produce a fluent paragraph citing "Section 124A of the Indian Contract Act" and you will not know — without independently checking — that no such Section exists.

This page walks through the three structural problems and explains where Clauseium addresses each.

Problem 1: Hallucinated citations

Large language models hallucinate citations by design. They generate text token-by-token based on statistical likelihood. When asked about a specific statutory Section, the model produces something that looks like a Section reference even if it doesn't exist.

For Indian law specifically, the problem is worse because the training data is sparser. ChatGPT has seen comparatively few Indian legal documents during training. When asked about an Indian statute, it interpolates from US/UK common-law patterns. The result: confident-sounding but wrong outputs.

The published research is sobering. A 2024 Stanford study found GPT-4 hallucinated legal citations in roughly 50% of US-law queries. For Indian law, the equivalent number is widely believed to be higher.

How Clauseium addresses this

Clauseium uses a retrieval-augmented generation (RAG) pipeline grounded in Indian Kanoon (16M+ judgments) and the live India Code. The pipeline runs three verification stages:

1. Extraction: when the model proposes a Section reference, the reference is parsed and structured.
2. Retrieval: the parsed reference is looked up against the live legal corpus.
3. Output gate: if the lookup fails — Section doesn't exist, or the cited text doesn't match — the citation is suppressed before it reaches the user.

The result: zero hallucinated citations in production output. If Clauseium can't verify a Section, it doesn't claim it.

For deeper context on indemnification clauses where citation accuracy matters most, see the Section 124 indemnification deep-dive.

Problem 2: No Indian-law grounding

ChatGPT is trained on a general web corpus. The corpus contains some Indian legal materials — Wikipedia summaries, news articles, blog posts — but does not contain the full India Code, the DPDP Rules 2025, the RBI Master Directions, or the SEBI regulatory framework as primary sources.

The result: ChatGPT can describe Indian law at the level of a general legal news reader. It cannot apply Indian law at the level of a practising advocate. The differences:

• Section 27 of the Indian Contract Act voids post-termination non-compete clauses. ChatGPT often misses this and applies US-style non-compete reasoning.
• The MSME Act 45-day payment rule overrides longer payment terms. ChatGPT treats payment terms as freely negotiable.
• DPDP Act Section 8(5) requires written processor contracts with sixteen specific obligations. ChatGPT treats data protection as a generic concept similar to GDPR.
• Stamp duty varies by state and contract type. ChatGPT often invents a single national rate.

For an in-house counsel relying on these analyses, the operational risk is significant.

How Clauseium addresses this

Clauseium's training and retrieval corpus is purpose-built around Indian law:

• Indian Contract Act, 1872 — every Section, with case-law cross-reference.
• DPDP Act, 2023 + Draft Rules 2025 — full text.
• IT Act, 2000 — including Section 43A, Section 79.
• Companies Act, 2013 — including SEBI overlays.
• FEMA, 1999 — for cross-border deals.
• Indian Stamp Act, 1899 + state stamp acts.
• Arbitration and Conciliation Act, 1996.
• 16M+ Indian Kanoon judgments for case-law grounding.

Every clause analysis cites these primary sources, and every citation is verified before display.

Problem 3: Confidentiality and DPDP

The third problem is operational, not technical. Pasting a contract into ChatGPT involves three actions that may not be DPDP-compliant:

1. Sharing personal data: contracts contain names, email addresses, sometimes KYC data. Under DPDP Section 2(t), processing personal data requires a lawful basis.
2. Engaging a data processor: ChatGPT, in this transaction, becomes a data processor under DPDP Section 2(u). Section 8(5) requires a written contract with the processor.
3. Cross-border transfer: Standard ChatGPT processes inputs in US data centres. DPDP Section 16 permits cross-border transfer subject to government notification of restricted countries; while no restrictions exist as of May 2026, sectoral RBI directions for BFSI customers may apply.

Standard ChatGPT does not provide the Section 8(5) processor contract. The standard product's terms include training on user inputs unless explicitly disabled. For an Indian fiduciary, pasting a contract into standard ChatGPT is a DPDP risk that should not be taken lightly.

How Clauseium addresses this

• Zero data retention: contracts are processed in-memory; no data is stored beyond the user's encrypted workspace. Foundation model providers (OpenAI, Anthropic) are engaged under zero-retention agreements.
• AWS Mumbai data residency: all processing in ap-south-1; no cross-border transfer.
• DPDP-compliant DPA: Clauseium signs a Section 8(5) processor agreement with every customer.
• SOC 2 Type II + ISO 27001 + ISO 42001: independently audited security controls.

For more detail on DPDP processor obligations, see the DPDP compliance guide.

When ChatGPT is genuinely useful

To be balanced: ChatGPT is a tool worth keeping in the legal toolkit, just not for the contract-review workload.

Good uses for ChatGPT in legal work:

• First drafts of non-binding prose: meeting summaries, internal memos, presentation outlines.
• Brainstorming: generating a list of risk categories or negotiation positions to consider.
• Document summarisation: for briefings, where the underlying document is the source of truth.
• Translation and language polish: improving clarity of internal communications.
• Generic legal research: as a starting point that you then verify against authoritative sources.

For these uses, ChatGPT is fast, cheap, and adequate. The output is checked by the user before it gets relied on.

Bad uses for ChatGPT in legal work:

• Contract review where outputs get relied upon without independent verification.
• Citing specific Indian-law Sections in client-facing analysis.
• Processing contracts containing personal data without DPDP-compliant infrastructure.
• Generating legal opinions that go to a regulator or counterparty.

The dividing line is: does the output get used as fact, or as input to your own thinking? ChatGPT is fine for the second; not safe for the first.

A practical decision framework

Use ChatGPT for:

• Brainstorming, summarisation, internal drafting.
• Non-binding research where you'll verify before relying.
• Workloads where citation accuracy doesn't matter.
• Personal productivity tasks.

Use Clauseium for:

• Production contract review.
• Client-facing analysis with Indian-law citations.
• DPDP compliance scanning.
• Any workload where the output gets relied upon.

Don't use either for:

• Final legal opinions on novel transactions — those still need a Bar Council-enrolled advocate's judgment.

Bar Council of India ethics consideration

A note for advocates specifically: the Bar Council of India Rules require advocates to maintain client confidentiality and exercise reasonable care. Pasting a client's contract into a general-purpose LLM that may train on the input and stores history in foreign data centres potentially conflicts with both obligations.

This is not yet a settled ethics question — there are no published Bar Council disciplinary cases on LLM use specifically — but the conservative reading is that advocates should use legal AI tools that are contractually committed to confidentiality and DPDP-compliant data handling. Clauseium meets that bar; standard ChatGPT does not.

Final verdict

ChatGPT is not a substitute for legal AI. It is a general-purpose assistant that is excellent at general-purpose tasks and structurally unsuitable for production legal work. The hallucination rate, lack of Indian-law grounding, and confidentiality posture make it unsafe for client-facing contract review.

Clauseium is purpose-built for the workload. Indian-law grounded, citation-verified, DPDP-compliant, audit-trail capable. The right tool for the job.

The trade-off is cost: Clauseium is paid; ChatGPT has a free tier. For any contract that will be relied upon — vendor agreement, customer MSA, employment contract — the price difference is rounding error compared to the litigation risk of acting on a hallucinated citation.

Try Clauseium free →