July 2, 2026, (Inside AI) — For three years, three police constables in Chhattisgarh’s Bastar region allegedly siphoned off nearly Rs 2 crore by inflating their own salaries. The scheme went unnoticed until government auditors deployed artificial intelligence tools to analyze payroll data for roughly 2,000 personnel. This marks the first known instance of AI being used for such an audit in the state’s police force.
The accused—Girish Rai, Rajkumar Katlam, and Hemant Mathew—were arrested this week. Rai, an assistant in the salary section of the Superintendent of Police (SP) office in Jagdalpur, allegedly manipulated soft copies of salary records before processing. By inflating figures by small amounts each month, the trio extracted between Rs 1.5 crore and Rs 2 crore from October 2023 to May 2026.
The fraud exploited the natural volatility of police payrolls. Transfers, postings, and fluctuating personnel strength create constant noise in expenditure data, masking the incremental theft. Auditors, facing a massive dataset, turned to AI for pattern recognition beyond human capability.
Bastar SP Shalabh Kumar Sinha confirmed the proactive use of technology:
“The data related to salaries was voluminous and so auditors proactively decided to use AI tools.”
An investigator explained why traditional methods failed:
“The fraud went undetected because police salary expenditure fluctuates frequently due to regular transfers, postings and changes in personnel strength. The accused allegedly inflated salaries by small amounts each month in their own names and those of the other two constables, allowing the fraud to remain unnoticed for years.”
When Spreadsheets Hide a Slow Bleed
Financial fraud in government payrolls often thrives on obscurity. In large organizations, minor discrepancies can hide inside thousands of line items. AI changes this equation by learning normal patterns and flagging deviations—even tiny ones—that persist over time. In this case, the tool likely detected unnatural consistency in specific employees’ pay bumps, something a human auditor might dismiss as error or overlook entirely.
The Bastar case mirrors a global shift toward algorithmic auditing. India’s Comptroller and Auditor General has been experimenting with AI for fraud detection in welfare schemes. State governments in Maharashtra and Karnataka have piloted similar tools for procurement and tax compliance. But applying it to internal police salaries is novel—and risky, given the sensitivity of personnel data.
Critics note that AI audits are only as good as the data they train on. If historical records contain hidden fraud, the model might learn to accept it as normal. Moreover, the “black box” nature of some algorithms can make it hard to explain why a particular flag was raised, complicating legal proceedings. In Bastar, however, the AI’s output reportedly aligned with manual verification, leading directly to arrests.
The three constables face charges of cheating, forgery, criminal breach of trust, and misappropriation of government funds under the Bharatiya Nyaya Sanhita (BNS). They were remanded to 14 days of judicial custody. The investigation continues to determine if more employees were involved or if similar fraud exists in other districts.
The Thin Blue Line of Trust
This breach strikes at institutional integrity. Police forces rely on internal discipline, yet here the fraud originated from the salary section itself—a unit meant to ensure fair compensation. The use of AI may deter future misconduct, but it also raises questions about oversight gaps that allowed three years of theft.
Chhattisgarh police are now considering expanding AI audits to other administrative areas like travel reimbursements and procurement. The state’s home department has reportedly asked for a detailed report on the methodology used, aiming to replicate it across departments. Such moves align with India’s broader push for digital governance, though they demand robust data protection safeguards.
For now, the Bastar case serves as both a warning and a proof of concept. Small, steady fraud can no longer count on human fatigue to stay hidden. As one auditor noted, the sheer scale of data that once protected the guilty now betrays them—provided someone is willing to let the machines look.
