How our AI works

What it is

Answers are generated by Anthropic's Claude, a large language model, guided by instructions we wrote for Indian legal information. Every answer is AI-generated — no human reviews your question before you get a response.

Known limitations

• AI can be wrong: it may cite an outdated provision, miss recent judgments, or misunderstand your situation.
• It cannot assess evidence, predict case outcomes, or represent you anywhere.
• It provides general information — never a substitute for a qualified advocate's advice on your specific matter.

Safety rules we enforce

• The assistant refuses to help with fraud, evidence fabrication, harassment, or evading law enforcement.
• For urgent situations (arrest, violence, threats to safety), it leads with emergency helplines before any legal explanation.
• Every answer carries a disclaimer that it is information, not advice.

Where answers come from

Answers are grounded in a source registry of Indian legal provisions (acts, sections, and related material) that we curate and version by hand. Each entry goes through a draft → pending-review → approved workflow before it can be cited; only one approved version of a given provision is live at a time. There is no fixed automated re-ingestion schedule today — sources are added and revised manually as gaps or changes are found, not on a dated cycle.

What the confidence indicator means

Each answer runs through automated checks — including whether it cites a real, approved source and whether that citation actually supports what the answer says — before you see it. The confidence indicator reflects the outcome of those checks, not a human's judgment. A low score, or no matched source at all, means treat the answer with more caution and verify independently or ask a lawyer, rather than that the answer is necessarily wrong.

If you flag an answer

The thumbs up/down under each answer, and the email below, record your feedback against that conversation. Today this is a log, not an automated re-review loop — flagging an answer does not currently trigger anyone to look at it in real time. We read reports and use them to prioritize source fixes, but don't treat a report as changing the answer you already received.

Governance & accountability

• We log which AI model produced each answer, so any problematic output can be traced and investigated.
• Usage and spending limits prevent abuse of the service.
• Your conversations are protected by row-level security — only your account can read them — and are automatically deleted after 12 months of inactivity.
• You can download or permanently delete all your data anytime from your account page.

Report a problem

Got a harmful, biased, or clearly wrong answer? Email support@juriniva.com with the question you asked — we read every report, though see "If you flag an answer" above for what that does and doesn't do today.