India AI Impact Summit 2026

ai-driven-enforcement-better-governance-through-effective-compliance-services

8 min read

TRACK:Safe and Trusted AI
SPEAKERS
  • Abhishek Kumar, Central Board of Direct Taxes
  • Avneesh Pandey, SEBI
  • Harssh A Poddar, Supdt. Of Police, Nagpur Rural
  • Martin Wilcox, Teradata
  • Mausam, IIT Delhi
  • Ram Ganesh, CyberEye
  • Ramesh Revuru, LTIMindtree
  • Ravi Agrawal, Central Board of Direct Taxes
  • Sanjib Sanghi, ICAI
  • Shashi Bhushan Shukla, Central Board of Direct Taxes
  • Survendu Pati, Reserve Bank of India
  • Varunesh Mishra, Central Board of Direct Taxes
SESSION OVERVIEW

AI is transforming enforcement by enabling smarter compliance, better public services, and more responsive governance. This symposium aims to bring together, law enforcement, academia, and industry experts to explore how AI can strengthen institutional capabilities, improve service delivery, and support effective compliance, while promoting responsible adoption within evolving governance frameworks.

VIDEO RECORDING

AI-Driven Enforcement: Better Governance through Effective Compliance & Services

Abstract

The symposium brought together senior policymakers, technologists, and enforcement officials to examine how artificial intelligence can make Indian governance more proactive, fair and citizen‑centric. The opening keynote outlined the upcoming Income‑Tax Act 2025 and the need for a technology‑first ecosystem. Subsequent industry‑focused talks demonstrated concrete AI‑driven products – from the CBDT’s Project Insight 2.0 and LTIMindtree’s “Bharat Varsha” platform to IIT‑Delhi’s research on multimodal AI for policing and Teradata’s risk‑analytics suite. The regulatory segment showcased the RBI’s “Mule Hunter” AI for detecting mule accounts, Maharashtra Police’s “Mahakrime OS” investigative co‑pilot, SEBI’s AI tools for market‑surveillance, and the CBDT’s AI‑enhanced taxpayer‑service programmes. The session closed with a vote of thanks emphasizing that AI is now an operational pillar of Indian enforcement.

Detailed Summary

Speaker: Ravi Agrawal, Chairman, CBDT

  • Welcomed delegates to the AI Impact Summit theme “Sarvajan Hitay Sarvajan Sukhaye – Welfare for All, Happiness for All.”
  • Emphasised that the Income‑Tax Act 2025 (effective 1 April 2026) will be rule‑driven and technology‑centric, reducing interpretative ambiguity and expanding tax certainty.
  • Stressed AI’s role in amplifying human capability, turning massive data into actionable insights, automating routine work, and enabling faster, smarter decisions.
  • Linked AI adoption to the “Manav” vision articulated by the Prime Minister – accountability, ethical governance, sovereign data, inclusive AI, and legitimacy.
  • Highlighted two core messages: human‑centric AI (people must drive AI, not the reverse) and practical implementation (focus on data integration, risk scoring, anomaly detection, multilingual nudges, and workflow automation).
  • Cited early results: 1.11 crore taxpayers filed updated returns after AI‑driven nudges, generating ₹8,800 crore in revenue; foreign asset disclosures added ₹9,000 crore in assets and ₹6,500 crore in income.

Key Insights

InsightImplication
New Act simplifies language & mandates rule‑based processesEnables deterministic AI models, fewer litigation triggers
AI can shift human effort from “routine” to “enhanced” tasksRequires up‑skilling of tax officials
Early AI pilots already yielded billions of rupees in additional revenueDemonstrates quick ROI for AI‑enabled compliance

2. Category 1 – Industry & Academia

2.1 Project Insight 2.0 – Scaling AI‑Enabled Taxpayer Services

Speakers: Abhishek Kumar (CBDT), Ramesh Revuru, Nivasan T (LTIMindtree)

  • Project Insight is the department’s long‑standing data‑warehouse & BI platform that aggregates internal tax data with third‑party sources.

  • Insight 2.0 adds AI/ML/LLM layers to deliver:

    1. Expanded Annual Information Statement (AIS) – broader coverage, richer data.
    2. Improved data correctness – end‑to‑end digital linkage lets taxpayers flag and correct third‑party errors.
    3. Conversational chatbot – LLM‑driven UI for line‑item queries, automatic escalation to grievance‑tracking when needed.
    4. Pre‑filled returns – more accurate pre‑population using enriched data and taxpayer feedback.
    5. Intelligent e‑verification – faster matching of returns with third‑party data, capable of ingesting unstructured inputs via LLMs.
    6. Dynamic, AI‑powered nudges – context‑aware, multilingual (covering major Indian languages), multichannel (SMS, email, in‑app).
    7. Litigation‑risk scoring – LLM‑tagged assessment/orders to predict case win‑ability and reduce litigation.

  • Anticipated outcomes: faster filing, higher compliance, lower false‑positive alerts, and a measurable drop in litigation volumes.

2.2 “Blueverse → Bharat Varsha” – Sovereign LLM Platform

Speaker: Ram Ganesh, CyberEye (presenting LTM’s offering)

  • Introduced Blueverse, an agentic platform for building domain‑specific AI agents without writing code.
  • Announced the India‑specific, “Bharat Varsha” version, purpose‑built for the CBDT.
  • Platform architecture consists of five pre‑built layers: foundational models, LLMs, data, knowledge, orchestration, and consumption.
  • Described a small‑language‑model (SLM) approach: fine‑tuning a generic LLM on a 1‑2 % parameter subset (using LoRA) with curated tax‑domain data, ensuring sovereign, secure, and low‑cost models.
  • Highlighted vector‑DB + retrieval‑augmented generation (RAG) to provide source citations, multilingual support, and legal‑interpretation intelligence.
  • Demonstrated a use‑case for auto‑detecting mule‑account patterns and real‑time transaction scoring.

2.3 AI for Law‑Enforcement – Research Perspectives

Speaker: Mausam, IIT Delhi

  • Presented a broad AI‑for‑law‑enforcement roadmap: prevention, prediction, investigation, and post‑crime analysis.

  • Illustrated multimodal AI (visual, textual, speech) enabling new use‑cases:

    • CCTV + facial‑recognition – reported 27 % crime reduction in Surat (2014).
    • Satellite‑imagery for maritime surveillance & border monitoring.
    • Anomalous vehicular behaviour detection (e.g., “car‑rape” case in Delhi).
    • Audio‑to‑text for rapid FIR drafting and chatbots for citizen assistance.

  • Emphasised human‑in‑the‑loop to avoid over‑triggering, bias, and loss of public trust.

  • Identified core challenges: data silos, algorithmic bias (especially across caste, gender, and region), and privacy‑civil‑liberty safeguards.

2.4 AI‑Driven Risk Analytics for Financial Crime

Speaker: Martin Wilcox, Teradata

  • Showcased in‑warehouse AI for anti‑money‑laundering (AML) at a large US bank: moving analytics to the data warehouse cuts cost, complexity and enables training on the full petabyte‑scale dataset.

  • Described ensemble models (supervised + unsupervised) for anomaly detection in a European retail bank, tackling the classic imbalanced‑class problem.

  • Demonstrated graph analytics for collusion detection in an Asian bank’s credit‑card fraud ring – stressing that graph‑based AI must run at scale (O(N²) challenge) and thus be executed inside the warehouse.

  • Introduced next‑generation multimodal AI:

    1. Income‑estimation models for Brazil’s un‑banked (25× faster inference when “brought‑your‑own‑model” into the warehouse).
    2. NPS drivers extraction from 50 k weekly chat logs (text + sentiment analysis).

  • Announced a forthcoming sovereign‑AI offering for Indian financial institutions.

3. Category 2 – Regulatory & Enforcement Bodies

3.1 RBI’s “Mule Hunter” – AI for Detecting Mule Accounts

Speaker: Survendu Pati, RBI

  • Recapped the RBI’s AI‑sandbox framework (7 sutras, 6 pillars) adopted by the Government of India on 5 Nov 2023.

  • Described Mule Hunter.ai:

    • Trained on 857 engineered features (bank‑specific subsets of ~50 high‑impact features).
    • Deployed on‑prem at 26 banks, with a central aggregation service for model updates while keeping raw data siloed.
    • Achieved 80‑90 %+ accuracy versus 20‑30 % for prior rule‑based systems.
    • Identified novel patterns (e.g., spikes in midnight transactions, dormant‑to‑active‑to‑dormant cycles).
    • Early pilots suggest ₹75‑100 crore could be saved per bank by freezing mule accounts at “day 0”.

  • Future roadmap: real‑time transaction scoring, integration with telecom‑mobile‑number risk registries, and a public‑good ecosystem for cross‑bank learning.

3.2 Maharashtra Police – “Mahakrime OS” Investigation Co‑Pilot

Speaker: Harsh A Poddar, SP, Nagpur Rural & CEO, MARVAL

  • Introduced Mahakrime OS, an AI‑driven “investigation co‑pilot” launched with Microsoft (2025) and inaugurated by CM Devendra Fadnavis & Satya Nadella.

  • Scope: four high‑complexity crime domains – cybercrime, large‑scale economic fraud, organized narcotics, and serious sexual offenses.

  • Functionalities:

    1. Ingest FIR & documents → generate SOP‑compliant investigation path (state‑specific and judiciary‑validated).
    2. Automated legal requests (telecom & forensic data) and open‑source intelligence (OSINT) profiling.
    3. Case diary generation, “victim‑assistance” actions (e.g., account unfreezing).
    4. Guided workflow with human‑in‑the‑loop for each investigative step.

  • Piloted across 467 cases with 233 officers trained; earned several state‑level governance awards.

  • Highlighted underlying AI techniques: large language models, graph neural networks for relational analysis, agentic AI for workflow orchestration, and big‑data analytics for evidence synthesis.

3.3 SEBI – AI Across Market‑Surveillance & Cyber‑Compliance

Speaker: Avneesh Pandey, Executive Director, SEBI

Four AI‑driven platforms were presented:

PlatformPurposeNotable Outcomes
RIDARReal‑time monitoring of mutual‑fund advertisements for regulatory complianceDetected hundreds of non‑disclosures & missing disclaimer issues
SudarshanMultilingual AI‑engine for detecting misleading financial content on social mediaFlagged ~1 lakh instances of misinformation; continuous monitoring in 8+ Indian languages
InfomergeEnd‑to‑end workflow intelligence for investigations (data collection → report generation)Standardised reporting; reduced manual variance; human‑review step retained
Cyber‑Compliance AuditAutomated parsing of security‑audit artefacts submitted by market participantsIdentified missing controls; deployed ensemble‑model guard‑rails to prevent “hallucinations”
  • Emphasised democratisation of AI: development teams are internal to business units, not just the IT department.
  • Highlighted SEBI’s AI‑enabled cyber‑resilience framework (three‑model redundancy to catch hallucinations).

3.4 CBDT – AI‑Powered Taxpayer‑Service & Nudge Programme

Speaker: Shashi Bhushan Shukla, Principal Commissioner, CBDT

  • Reviewed the tax data ecosystem: > 80 crore PANs, > 12 crore ITRs, 650 crore SFT fields, plus CRS/FATCA filings (~5 million records annually).

  • Described the Saksham Nudge initiative (7‑step cycle: Sankalan → Anushandhan → C.R.I.A.N → ASTAC → Adhikar) that uses behavioural insights to pre‑emptively guide taxpayers.

  • Impact highlights (2022‑2024):

    • Foreign‑asset nudge (Dec 2023): 1.57 lakh taxpayers disclosed ₹99,000 crore of previously hidden assets → ₹6,540 crore additional tax.
    • Bogus‑donation deduction nudge: 6.96 lakh taxpayers withdrew ₹9,879 crore of illegitimate deductions → ₹1,758 crore extra tax.

  • Future roadmap: real‑time AI prompts at filing time, integration of AI‑driven litigation‑risk scores, and continuous feedback loops to refine nudges.

4. Closing Vote of Thanks

Speaker: Mahadevan K., Joint Commissioner, Income Tax (Delhi)

  • Re‑affirmed that AI is now operational, not aspirational, across tax, banking, securities, policing and central banking.
  • Thanked all speakers, moderators, and the organizing team for a “forward‑looking” dialogue.

Key Takeaways

  • Policy‑level vision: The Income‑Tax Act 2025 is deliberately rule‑driven and AI‑ready, providing a deterministic backbone for large‑scale automation.

  • Early ROI: AI‑driven nudges have already generated ₹8,800 crore in revenue and uncovered ₹9,000 crore in foreign assets, proving the financial upside of AI‑enabled compliance.

  • Sector‑wide AI platforms:

    1. Project Insight 2.0 – AI‑enhanced AIS, chatbots, pre‑filled returns and litigation‑risk scoring.
    2. Bharat Varsha (Blueverse) – Sovereign, low‑cost LLM architecture that can be fine‑tuned on tax‑domain data.
    3. Mule Hunter – Multi‑bank AI model achieving > 80 % detection accuracy for mule accounts.
    4. Mahakrime OS – AI co‑pilot that codifies SOPs, automates legal requests and fuses OSINT for complex investigations.
    5. SEBI’s suite (RIDAR, Sudarshan, Infomerge, Cyber‑Audit) – End‑to‑end AI for market surveillance, misinformation detection and security compliance.
    6. Teradata risk‑analytics – In‑warehouse AI for AML, anomaly detection, and graph‑based collusion identification at petabyte scale.

  • Human‑in‑the‑loop principle is a recurring safeguard: AI surfaces leads, but human analysts validate to preserve trust, mitigate bias, and respect civil liberties.

  • Data sovereignty & privacy are embedded in all initiatives (on‑prem deployments, anonymised cross‑institutional sandboxes, RBI’s AI‑sandbox policy).

  • Multimodal AI (combining text, image, audio, and structured data) is emerging as a game‑changer for law‑enforcement (e.g., satellite surveillance, CCTV‑based facial recognition, OSINT profiling).

  • Behavioural‑nudges backed by AI have shown measurable tax‑compliance uplift (foreign‑asset and bogus‑donation cases).

  • Future focus: Real‑time AI prompts at filing, cross‑jurisdictional data sharing, continuous model refinement, and scaling of sovereign LLMs across government agencies.

Prepared from the verbatim transcript of the AI‑Driven Enforcement symposium held in Delhi, 24 February 2026.

See Also:

Graph View

Backlinks