Tech-tonic shift: Bengal Police adopt an AI bot to tighten procedures and speed up probes
Criminals are using AI. Bengal Police are answering in kind. Across eight units, roughly 400 investigators - from assistant sub-inspectors to senior leadership - are now using an in-house AI bot to cut procedural errors and move cases faster.
Built by Birbhum Police with support from a Pune-based firm, the app runs on a closed corpus of Indian materials. It compiles 50,000+ pages: statutes, special laws, Supreme Court and High Court rulings, NHRC and MHA guidelines, police SOPs, training manuals, and standard responses. It was rolled out after months of trial and is not available on public app stores.
How it works on the ground
Enter one or two lines about a case, and the bot suggests applicable legal sections, procedural steps for the IO, and next actions for the victim. SP (Birbhum) Amandeep and SDPO (Bolpur) Rickey Agarwal led the build, with a clear brief: accelerate routine legal reasoning without diluting accountability.
In practice, officers report fewer drafting errors in case diaries and chargesheets. One officer-in-charge used the guidance to direct a complainant to file at the National Cyber Crime Reporting Portal after a KYC scam - exactly the step the law expects.
National Cyber Crime Reporting Portal
Guardrails from senior officers
Leadership has cautioned investigators: this tool does not detect crimes and it does not replace human sources. Over-reliance is a risk. The Directorate has set up an AI Cell to oversee usage, updates, and guardrails.
Why this matters to legal professionals
- Consistency of sections applied: Expect more uniform use of CrPC, IPC/BNS, IT Act, POCSO, and special laws across units. That improves predictability for both prosecution and defense.
- Discoverability and disclosure: Defense may seek the prompts, outputs, model version, and update history that informed an IO's decisions. Clarify what sits in the case diary versus what is administrative background.
- Admissibility hygiene: Any digital printout relied upon in proceedings should carry a Section 65B certificate. Treat AI suggestions as aides for drafting; the factual and legal basis must still be independently established.
- Due process and privacy: Inputs should be minimal, relevant, and logged. Avoid free-text entry of sensitive personal data. Map usage to constitutional privacy standards and departmental SOPs.
- Attribution and traceability: Every recommendation should point back to a statute, rule, or judgment. Outdated or uncited suggestions should not drive action.
- Accountability stays human: The IO, SHO, and prosecutors remain responsible for choices made. Notes on why a recommended section was accepted or rejected will matter in court.
Practical checks for police departments rolling out similar tools
- Scope control: Restrict use to official devices and networks. Disable external sharing by default.
- Immutable logs: Time-stamped prompts and outputs, tied to user IDs. Preserve logs with retention periods aligned to limitation statutes and departmental policy.
- Legal vetting loop: A review cell (law officers + senior IOs) should audit sampled outputs weekly and issue corrections or updates.
- Citation-first outputs: Force the bot to show sources with paragraph or section references. No source, no action.
- Update protocol: Monthly corpus refresh with change logs and version numbers. Freeze versions per case to preserve auditability.
- Training and sign-offs: Short, scenario-based sessions for ASIs to SPs. Written acknowledgment that AI is advisory, not directive.
- Data minimization: Redact names, numbers, and bank details unless essential for the query.
What courts and defense counsel can ask
- Did the IO use the AI bot? If yes, which version and on what dates?
- What sections or procedures did the tool suggest, and what did the IO finally adopt? Reasons recorded?
- Are the cited authorities traceable to the statute, rule, or judgment text relied upon?
- Were digital records produced with proper certification and chain of custody?
Key use cases seen so far
- Faster section mapping for first reports and notices.
- Checklists for seizure, search, arrest, and 41A/41B compliance.
- Standard steps for cyber, financial, and online impersonation complaints, including victim guidance and portal routing.
- Cleaner, consistent drafting of case diaries and chargesheets.
Policy watch
If overseen well, this can cut routine mistakes (wrong sections, missed notices, late intimation) that derail trials. The AI Cell's job is to keep sources current, prevent drift, and keep the human decision-maker front and center.
Bottom line
AI can speed legal reasoning and surface checklists. It cannot replace investigation, judgment, or accountability. With tight logging, strict sourcing, and disciplined use, the net result is fewer procedural slips and quicker help for victims.
Further reading and upskilling
Your membership also unlocks:
