AI Use & Limitations

Honest information about what our AI does, what it cannot do, and how to use it safely.

Last updated: May 2026

1. What our AI does

Legal Setu uses large language models — currently OpenAI's GPT family — trained on a broad corpus of text including Indian statutes, leading case law, government portals, and standard legal commentary. The AI is configured to:

• Explain Indian legal concepts in plain language across English, Hindi, Marathi, Tamil, Telugu, Bengali, Kannada, and Gujarati
• Identify the likely applicable law, forum, and procedure for a given situation
• Cite specific Acts, Sections, and statutory provisions where relevant — including the new BNS, BNSS, and BSA which replaced IPC, CrPC, and Evidence Act with effect from 1 July 2024
• Draft letters, notices, complaints, and document templates based on standard clauses
• Prepare a structured case brief that the assigned advocate can review before your consultation

2. What our AI is not

The AI is not a substitute for a qualified advocate. Specifically, it is not:

• An enrolled member of any State Bar Council
• Authorised to provide legal advice within the meaning of the Advocates Act, 1961
• Able to appear before, file documents with, or represent you before any court, tribunal, or authority
• A source of binding legal opinion
• Capable of exercising professional judgment in the way a human lawyer does

3. Known limitations

You should be aware of the following inherent limitations of AI-generated legal content:

• Currency: The model's training data has a cutoff date. Statutory amendments, new notifications, and recent judgments after that date may not be reflected. We update prompts and references regularly, but the underlying model is not real-time.
• Hallucination: Large language models can occasionally generate plausible-sounding but inaccurate content — including incorrect section numbers, fabricated case citations, or wrong procedural details. We design our prompts to reduce this risk, but it cannot be eliminated entirely.
• State variation: Indian law varies significantly by state — stamp duty rates, court fees, rent control provisions, registration requirements, and limitation periods all differ. Where the AI generalises across India, it may not reflect your state's specific position.
• Fact-sensitivity: Legal outcomes depend heavily on specific facts, documentary evidence, and procedural posture. The AI sees only what you describe, in the words you describe it. It may misjudge facts you consider minor or miss facts you have not mentioned.
• Language nuance: Translations across Indian languages, while careful, may lose statutory nuance. Where precise legal wording matters — for example, in a notice or complaint — the English statutory text governs.

4. When to confirm with a lawyer

Confirm with a qualified advocate before acting on any AI-generated content if your matter involves:

• Any court, tribunal, or quasi-judicial proceeding
• A criminal complaint, FIR, arrest, summons, or warrant
• A financial dispute above a few thousand rupees, or any matter involving fraud, breach of trust, or recovery
• Real estate transactions (purchase, sale, mortgage, gift, partition)
• Marriage, divorce, child custody, maintenance, or domestic violence
• Employment termination, retrenchment, or industrial disputes
• Wills, succession, and estate administration
• Company or partnership formation, dispute, or dissolution
• Any matter where you face a statutory deadline (limitation period, appeal window, notice period)

For these matters, the AI is useful for preparing your understanding before the consultation. The advice itself should come from a qualified advocate.

5. How we mitigate AI risk

We take several steps to reduce the likelihood of harmful AI errors:

• Prompt design: Our prompts instruct the model to cite specific Acts and Sections, flag uncertainty, recommend professional consultation for serious matters, and use the post-July-2024 BNS/BNSS/BSA references for new criminal matters
• Editorial review: Blog content, document templates, and key prompt instructions are reviewed by human editors familiar with Indian law before publication
• Confidence cues: Where the AI is unsure, it is instructed to say so rather than fabricate a confident answer
• Lawyer escalation: Every paid consultation includes a human advocate who reviews the AI-prepared case brief before speaking to you
• Feedback loop: User-reported errors are reviewed and used to refine prompts and templates

6. What we don't do with AI

• We do not use AI to generate advice attributed to a specific lawyer
• We do not use AI to replace the lawyer in a confirmed consultation
• We do not represent AI output as the equivalent of professional legal opinion
• We do not sell your chat content to advertisers or third parties — see our Privacy Policy
• We do not train external public models on your private chat content (model providers process your queries to generate responses under contractual data-protection terms; they do not use the content for general model training)

7. Your responsibilities

When using the AI, you are responsible for:

• Providing accurate and complete information
• Verifying AI-generated content before acting on it for any consequential matter
• Paying the correct stamp duty, registration fee, or notarisation cost on any generated document
• Seeking professional advice for the categories listed in section 4
• Not using the platform to seek advice on conduct that is itself unlawful

8. Reporting AI errors

If the AI gives you content that you believe is materially wrong — particularly an incorrect statutory citation, a wrong procedural step, or an out-of-date reference — please report it to hello@legalsetu.co.in with the subject line "AI Error". Include a screenshot or the exact text. Verified errors are corrected and added to our prompt refinement process.

9. Contact