Reports indicate that Indian citizens lost around ₹22,000 crores due to cyber fraud last year. And the last thing the country needs, as AI makes it easier than ever to produce deepfakes and synthetic identities, is another Jamtara like situation.
Called as the phishing capital of India, the Jharkhand district’s unfortunate association with phone-based financial fraud has made its name synonymous with the scam industry.
Today, multiple tech companies are tackling the challenge head-on, and one such is Bureau.
The company aims to prevent new fraud hotspots from emerging while helping places like Jamtara shed their infamous reputations.
Why Bureau Exists
Founded in 2016, the company fights fire with fire — by using AI to combat AI-enabled fraud. The company lists leading firms in India, such as Swiggy, Tata, Rapido, and Jar, as well as prominent banks like IDFC First, among its customers.
“Fraud has become a factory,” said Venkat Srinivasan, the chief analytics and risk officer at Bureau, highlighting the increasing number of fraud clusters and networks across countries in South Asia.
“What really should be a factor [in detecting these fraud networks] is identifying the connections,” said Srinivasan. “If you look at the data individually, you may not find anything wrong.”
He said that each individual may share the same identities, devices, phones, IP addresses, and email addresses, and transact with one another. “The best way to bring them all together is a graph, and we use it to find strong linkages,” added Srinivasan.
For example, the system might identify a single telephone number associated with two different PAN cards, which in turn are linked to 10 other devices. This is a red flag indicating coordinated fraud. Such patterns are only visible when analysed through the graph network’s perspective.
And when such identities are linked, it may result in a “village” of 200 fraudsters. This aids in creating a strong network graph, which forms the foundation of Bureau’s Graph Identity Network.
The company’s approach, which utilises graph networks and advanced AI models and algorithms, has resulted in a 95% reduction in collusion-based fraud for its clients.
Besides, Bureau has also integrated mechanisms that help effectively segment false positives, or ‘good users’, from fraud rings based on shared traits, industries, and linkages.
Targeting Mule Operations
These graph-based solutions also help identify mule operations and networks. Money mules identify individuals with weak digital identities, including accounts where names don’t match across platforms, incomplete authentication records, and social footprints that appear fabricated.
Bureau’s algorithms and systems help their customers stay compliant with anti-money laundering regulations by assessing a Mule Score.
“Our solutions are designed for people with fragile digital profiles who haven’t gone through robust authentication processes,” said Srinivasan.
These are typically individuals whose identity documents remain unverified against one another, who don’t appear in employment databases like EPFO, have no tax records, such as GST registrations, and lack an established digital presence.
The company said that the deployment of the Money Mule Score with a leading Indian bank led to a 60% uplift in money mule detection as compared to the bank’s existing KYC process.
“These early mule detections enabled the bank to prevent the potential fraud losses of over $43 million within the first six months of using the solution,” said the company in a statement.
In another case, a Tier One bank in Southeast Asia was experiencing significant fraud losses from new savings accounts that passed KYC tests but showed rapid money movement and defaults.
Bureau analysed 600,000 applications over two months, which led to a 58% increase in mule detection accuracy, preventing approximately $2.1 million in downstream fraud. The approach helped reduce false positives by 30%, the company stated.
Magic of Behavioural Biometrics
Another intelligence layer built into the platform is the company’s ability to prevent fraud using ‘behavioural biometrics’. Sandesh G S, the company’s CTO, explained how it works in preventing sophisticated account takeover attacks that combine stolen credentials, SIM swaps, and rapid fund transfers.
A fraudster buys stolen usernames and passwords on the dark web, trying many until some work. He then uses SIM swapping to control messages and OTPs, quickly adding beneficiaries and transferring funds to his network.
“If you’re working with Bureau, we can identify how the fraudsters are entering their password,” said Sandesh.
He explained how Bureau’s solutions, when integrated with any banking or financial platform, can analyse keystrokes to assess the associated risk. It is capable of noticing a difference between how a legitimate user enters their password and how a fraudster attempts to gain illegal access to the same account.
The Bureau then provides a risk indicator when the password entry shows an unusually low similarity score, suggesting that the user may be unfamiliar with the credentials or that the account has been compromised. This prompts the platform to initiate additional verification and authentication measures.
Sandesh also highlighted that the company offers APIs capable of detecting SIM swap fraud while simultaneously monitoring patterns such as the frequency and volume of beneficiaries being added to an account.
To Fight Fire With Fire
Bureau’s solutions are increasingly relevant as banking and financial institutions show a rising interest in AI and agentic technologies.
Industry experts have raised concerns about associated risks and emphasised the need for additional safeguards. Thus, the adoption and propagation of AI should not lead to downstream risks.
In a press conference at the Splunk .conf25 event, Ryan Fetterman, senior manager of AI security research at Cisco, said “We have to remember that attackers are also on an adoption curve. And as much as we’re trying to figure out the natural fit for AI solutions on defence, they’re trying to do the same things on offense.”
Rishi Aurora, managing partner at IBM Consulting India and South Asia, said that beyond known issues like hallucinations and data bias, “Other areas of concern include cybersecurity risks, data leaks and unauthorised access that expose sensitive information.”
“To mitigate these risks, agentic AI algorithms should have access controls and authentication mechanisms to prevent unauthorised interactions,” he said.
Enterprises need to move away from legacy systems and processes to integrate AI solutions that help mitigate such risks, Aurora noted.
Citing IBM as an example, he mentioned the ‘Pillars of Trust’ framework that the company implements while building AI products and solutions at scale. The framework includes explainability, fairness, robustness, transparency, and privacy.
Such frameworks exemplify efforts to develop responsible AI systems that avoid abuse — the very challenge that companies like Bureau are tackling head-on.
The post How to Prevent Another ‘Jamtara’ appeared first on Analytics India Magazine.