MahaCrimeOS AI: Maharashtra Police's generative-AI investigation platform

Cybercrime complaints in India have risen steeply. A Microsoft feature on the project cites I4C data, quoted in a parliamentary reply of 2 December 2025, of about 2.27 million cybercrime cases in 2024 and losses of more than 228 billion rupees.² Maharashtra's response was to create MARVEL, the Maharashtra Advanced Research and Vigilance for Enforcement of Reformed Laws. The Home Department's page records that the state approved the special purpose vehicle in principle by a decision of 16 February 2024 and gave administrative approval by a Government Resolution of 16 March 2024. MARVEL is a private limited company whose share capital is entirely held by the Government; the Superintendent of Police, Nagpur (Rural), is ex-officio director and CEO reporting to the Home Department, the Director of IIM Nagpur is an ex-officio director, and its objective is to help the state police enforce the law more effectively using AI.³

On 10 March 2025 the state's Finance Minister announced in the budget speech an AI-powered command and control centre at the Mumbai Police Commissioner's office for crime prevention, emergency response and public safety, without explaining which crimes it would target or giving a cost; MediaNama raised concerns about biased training data and noted that, unlike the EU AI Act, India has no restriction on AI assessing the risk of individuals committing offences.⁴

According to Microsoft and MediaNama, MahaCrimeOS AI is a generative-AI platform built by CyberEye, a Microsoft partner that worked with MARVEL and the Microsoft India Development Center, on Azure OpenAI Service in Microsoft Foundry and secured through Microsoft Defender for Cloud.¹² Its reported functions are:¹²

• extracting information from unstructured inputs such as complaints, chat messages, bank records, screenshots and handwritten notes, in English, Hindi and Marathi, and building a case file;
• drafting FIRs and official requests to banks, telecom companies and online platforms, including requests to freeze accounts, block numbers or take down accounts, and tracking the replies;
• providing built-in access to India's criminal laws and open-source intelligence, extracting crime identifiers and listing applicable laws;
• suggesting next steps such as analysing call detail records or gathering OSINT, and producing tailored investigation plans based on Maharashtra police protocols;
• a dashboard showing each investigator's active cases, pending actions and responses, and storage of case diaries usable in court.

MediaNama's analysis stresses that the tool influences investigative decisions, not just paperwork, because suggestions on next steps and legal sections shape what officers do.¹

Nagpur district police have used MahaCrimeOS AI since April 2025. On 12 December 2025 the Maharashtra government and Microsoft announced that it would expand from 23 police stations in Nagpur to all 1,100 stations statewide.² Microsoft's feature quotes Nagpur police officers on the effect: an Assistant Police Inspector said that collecting information for a case, including replies from service providers, had gone from two to three months to about a week, and that he now handles seven to eight cases a month instead of one; an Assistant Investigator said her caseload rose from two or three cases a month to about 200 a month over two years; and registering an FIR for one bank clerk's case took 15 minutes.² The feature also reports Chief Minister Devendra Fadnavis saying that 'ethical and responsible AI for public good is our core motto'.² These figures come from users and the vendor and have not been independently evaluated. MediaNama's piece dated 17 December 2025 reported the statewide rollout, and the Microsoft feature gave 12 December as the announcement date.¹²

MediaNama's analysis, 'What Maharashtra's AI policing rollout means for investigations, bias and fair trials', set out several concerns.¹

• Evidence law and fair trials. It is unclear whether AI outputs are admissible, who could be cross-examined about them, and who would qualify as an expert witness.
• Bias. Historical crime data and predefined categories could reinforce existing biases.
• Accuracy. Hallucinations and misreadings could misdirect investigations or delay actions such as account freezes, and errors may carry through a case.
• Over-reliance. Officers under pressure may defer to AI outputs without scrutiny, which makes training essential.
• Governance and data. Authorities have disclosed little about the datasets used, data retention periods or independent bias audits, and it is unclear whether standard operating procedures exist for adoption, monitoring and model drift. Privacy concerns centre on the sharing of large volumes of personal data between state agencies and private companies.

The article quotes Dona Mathew of the Digital Futures Lab as saying that AI in law enforcement needs to be preceded by an assessment of potential risks and that operationalising responsible AI should not be left to developers alone.¹ The state's rationale is capacity: rising complaints and the need to coordinate with many banks, telecoms and platforms.¹ MediaNama's coverage of Hyderabad police's plans to use AI in investigations raised parallel questions about accountability between developer and police, and about precedents such as the 2020 AI-based mask enforcement on CCTV.⁵

India has no statute that specifically regulates police use of AI. How the Digital Personal Data Protection Act, 2023 applies to this processing is not addressed in the sources reviewed for this entry.

As of 15 October 2026 the sources reviewed show MahaCrimeOS AI as announced for statewide use following the December 2025 announcement. I could not verify how many stations have been connected, whether any evaluation, audit or SOP has been published, or whether any court has examined evidence generated with it. MediaNama's December 2025 analysis remains the main published scrutiny, and the project's funding was not stated in the sources.¹²