India’s banks enter the new financial year from a position of unusual balance-sheet strength. Bad loans are near historic lows, capital buffers remain comfortable and regulatory stress tests suggest that the banking system can withstand even severe economic shocks.
Yet the threat most troubling financial institutions may not appear on a loan book or market-risk dashboard.
It could arrive as a perfectly worded email from a supposed bank executive, a phone call in the unmistakable voice of a family member, a realistic video of a senior official authorising a payment, or malicious software capable of changing its behaviour faster than conventional security tools can detect it.
In its Financial Stability Report released on June 30, 2026, the Reserve Bank of India identified artificial intelligence-enabled cyberattacks as the leading near-term cyber risk perceived by banks and non-banking financial companies. The assessment reflects a significant change in the nature of financial crime: attackers are no longer using AI merely to improve old scams. They are using it to industrialise deception, automate attacks and target thousands of victims simultaneously.
“The danger is not simply that AI makes cybercrime more sophisticated. It makes convincing fraud cheaper, faster, more personalised and far easier to scale.”
The RBI’s warning does not mean that India’s banking system is on the verge of collapse. The same report found that scheduled commercial banks remain financially resilient, with gross non-performing assets at about 1.8% in March 2026 and expected to remain below 2% through March 2028 under the baseline scenario. But financial resilience and digital resilience are different questions. A bank may have strong capital and still suffer a damaging technology outage, data breach, payment fraud or loss of customer confidence.
From badly written emails to AI-personalised phishing
Traditional phishing often depended on crude messages containing spelling mistakes, generic greetings and suspicious links. Generative AI has removed many of those warning signs.
Attackers can now create polished emails in English, Hindi and regional languages, reproduce the tone of a bank, employer or government department, and customise messages using information gathered from social media, leaked databases and previous breaches.
A customer may receive a message that appears to know the name of the bank, the type of account held, the city in which the person lives and even a recent purchase. The communication may claim that KYC has expired, a credit card has been blocked, reward points are about to lapse or an unauthorised UPI transaction has been detected.
The objective remains familiar: persuade the customer to click a link, install an application, share an OTP, enter a UPI PIN or disclose credentials. What has changed is the credibility and volume of the approach.
AI tools can generate thousands of variations, test which messages produce responses and adapt future attempts. Fraudsters can also use chatbots to maintain extended conversations with victims, answering questions in real time and reducing the inconsistencies that once exposed human-run scams.
“The age of spotting phishing through poor grammar is ending. A fraudulent message can now be as polished, contextual and persuasive as a genuine bank communication.”
Financial institutions therefore face a dual challenge. They must filter malicious communication before it reaches employees and customers, while also educating users that professional language, accurate personal details and familiar branding no longer prove authenticity.
Voice cloning turns trust into an attack surface
Voice-cloning fraud is among the clearest examples of AI converting human trust into a security vulnerability.
Audio samples collected from social-media videos, interviews, public speeches or voice notes can be used to imitate a person’s voice. A criminal can then pose as a relative requesting emergency money, a senior executive ordering an urgent transfer, a bank representative verifying a transaction or a police officer threatening immediate legal action.
In one reported Indian case, an Indore resident lost ₹1.83 lakh after fraudsters allegedly used an AI-generated voice to impersonate a relative living abroad. The attackers claimed that the relative faced a visa emergency and needed money immediately. The deception became clear only after the victim contacted the real person separately.
Research published in 2026 underlines why this method is dangerous. In a controlled study of synthetic and human voices used in vishing-style scenarios, participants were unable to distinguish them reliably; most AI-generated clips were judged to be human. The findings suggest that relying on tone, pauses or emotional expression is no longer an adequate defence.
The practical lesson is simple: a familiar voice should not be treated as authentication.
Families and businesses increasingly need secondary verification methods. An unexpected request for money should be confirmed by calling the person on a previously saved number, asking a question known only to the parties involved or using a pre-agreed family codeword.
Within banks and companies, high-value payment instructions should never be approved solely through a voice call, video conference or messaging application, regardless of how convincing the executive appears.
Deepfake video and synthetic identity fraud
Voice cloning is only one part of the problem. AI-generated images and videos can be used to impersonate senior executives, customers, government officials and law-enforcement personnel.
These capabilities can strengthen “digital arrest” scams, in which victims are falsely told that they are under investigation and must transfer money for verification or safekeeping. They can also be used to bypass remote onboarding systems, manipulate video-KYC processes or create synthetic identities assembled from a mixture of real and fabricated information.
India’s cybercrime authorities have warned that compromised identity-verification systems could allow criminals to complete fraudulent KYC processes, activate digital wallets, open accounts or gain unauthorised access to existing financial services. Such accounts may subsequently be used as mule accounts to receive and rapidly redistribute stolen funds.
This creates risk beyond the immediate victim. Fraudulent accounts can become part of a laundering chain, moving money across several banks, wallets and payment platforms within minutes. By the time a customer realises what has happened, the funds may already have been divided into smaller amounts and transferred through multiple intermediaries.
Automated malware raises the speed of attack
AI is also changing malicious software.
Cybercriminals can use generative tools to help write scripts, identify software vulnerabilities, create convincing malicious documents and modify malware so that it is harder for signature-based security systems to recognise. Less-skilled attackers can obtain technical assistance that would previously have required experienced programmers.
More advanced threats may use automation to scan large numbers of systems, identify exposed services, generate tailored attack code and alter their behaviour after encountering security controls.
For banks, the target is not limited to customer accounts. Attackers may go after employee credentials, payment-processing systems, cloud environments, software suppliers, call centres, fintech partners and third-party service providers.
A successful intrusion could result in data theft, fraudulent transfers, ransomware, service disruption or manipulation of information. Even when customer deposits remain safe, an extended outage can affect payment flows, market operations and public confidence.
The interconnected structure of modern finance increases the stakes. Banks rely on cloud providers, telecom networks, outsourced technology firms, payment gateways, fintech platforms and shared digital infrastructure. A vulnerability in one supplier may provide a route into several institutions.
“Cyber resilience is no longer determined only by the security of a bank’s own servers. It depends on the weakest critical link across its entire technology and vendor ecosystem.”
Why India is particularly exposed
India’s digital payments transformation has brought enormous benefits. UPI, mobile banking, Aadhaar-enabled services and low-cost digital accounts have made financial transactions faster and more accessible.
The same scale, however, offers criminals a vast pool of potential targets.
A successful fraud method can be deployed across millions of mobile numbers at low cost. India’s linguistic diversity, large population of first-time digital users and rapid adoption of app-based financial services create opportunities for highly localised scams.
AI can generate messages and audio in regional languages, imitate local accents and tailor stories to cultural or family contexts. Fraudsters may combine data from leaked databases, social platforms and public records to make their approaches appear credible.
The psychological techniques remain familiar: urgency, fear, authority, greed and emotional pressure. AI simply makes them more convincing.
Customers are told that an account will be frozen, a parcel contains illegal goods, a family member has been arrested, a refund is waiting, an investment opportunity will expire or a payment must be approved immediately. The victim is pushed to act before checking the story independently.
Are Indian banks ready?
India’s financial system is not starting from zero. Banks have invested heavily in security operations centres, encryption, multifactor authentication, behavioural analytics, transaction monitoring and fraud-detection systems.
The government, regulators, telecom operators and payment networks have also been building mechanisms to share intelligence and interrupt suspicious transactions.
The National Payments Corporation of India has reportedly begun piloting an AI-based system intended to trace fraudulent funds as they move across multiple accounts in real time. This is important because the ability to follow and freeze money quickly can be more valuable than identifying the original fraudulent transaction after funds have disappeared.
Large banks are also strengthening internal AI capabilities. HDFC Bank, for example, has disclosed work on a proprietary AI platform and enhanced fraud-monitoring systems as financial institutions seek more control over how sensitive banking data is analysed.
Nevertheless, the RBI’s assessment indicates that significant gaps remain. Technology alone will not be sufficient where employees can be socially engineered, vendors are inadequately monitored or institutions respond too slowly to emerging attack patterns.
Banks must also ensure that their own AI deployments do not introduce new vulnerabilities. Public generative-AI tools may expose confidential data if employees paste customer information, source code or internal documents into unapproved platforms. AI models used for threat detection must themselves be monitored for manipulation, bias, inaccurate alerts and unauthorised access.
Recent practitioner research has highlighted concerns including uncontrolled employee use of public AI tools, weak integration of AI into security operations, insufficient understanding of adversarial tactics and limited audit evidence around AI models used in financial cybersecurity.
What financial institutions must do next
The first priority is to move from reactive fraud detection to continuous, real-time risk analysis.
Banks should analyse not only passwords and OTPs but also device identity, location anomalies, typing behaviour, transaction history, beneficiary age, SIM changes, call patterns and unusual account movement. A transaction that appears valid in isolation may become suspicious when assessed against a wider behavioural profile.
Second, institutions must strengthen controls around high-risk actions. Adding a new beneficiary, increasing a transaction limit, changing a registered mobile number or transferring an unusually large amount should trigger additional verification and, where appropriate, a cooling-off period.
Third, banks need stronger deepfake-resistant identity checks. Video KYC and remote authentication should include active liveness tests, random prompts, document validation, device intelligence and cross-verification across trusted databases. A static selfie or apparently live face should not be considered sufficient.
Fourth, vendor security must become a board-level issue. Banks should maintain an inventory of critical third parties, assess their cyber controls, restrict access according to need and test how quickly services can be restored if a supplier is compromised.
Fifth, institutions must conduct simulations that reflect current threats. Employees should be tested against realistic AI-generated phishing, voice-cloned executive requests and deepfake video calls—not merely generic cybersecurity quizzes.
Finally, banks need rapid, coordinated incident response. Cybersecurity teams, fraud units, customer-service departments, legal teams, payment operators and law-enforcement agencies must be able to exchange information without bureaucratic delay.
What customers should do
Customers remain the final line of defence, particularly in scams that manipulate a person into voluntarily authorising a payment.
A few rules can significantly reduce risk:
Never share an OTP, card PIN, CVV, internet-banking password or UPI PIN. A genuine bank employee does not need these details to receive or reverse a payment.
Do not install screen-sharing or remote-access applications at the request of an unsolicited caller.
Do not approve a UPI collect request merely to “receive” money. Entering a UPI PIN authorises a debit.
Independently verify urgent requests from relatives, employers or officials using a known phone number.
Treat caller ID, profile photographs, voices and video appearances as potentially fake.
Avoid clicking links in unexpected KYC, reward-point, refund, parcel or account-blocking messages.
Keep banking applications, mobile operating systems and antivirus protections updated.
Enable transaction alerts and review account statements regularly.
Use lower transaction limits where possible and maintain a separate account for everyday digital payments.
Discuss voice-cloning and digital-arrest scams with elderly or less digitally experienced family members.
A household codeword can provide an additional safeguard against emergency voice scams. It should not be disclosed online or used in ordinary conversation.
What to do immediately after a fraud
Speed is critical.
Victims of cyber financial fraud in India should immediately contact their bank to block the account, card or digital channel involved. They should also call the national cybercrime helpline 1930 and file a complaint through the National Cyber Crime Reporting Portal.
The portal’s financial-fraud reporting system is designed to alert banks, wallets and financial intermediaries so that disputed funds may be identified and blocked before they move further through the system.
Victims should preserve screenshots, phone numbers, email headers, transaction references, account details, chat records and recordings. They should not delete messages even when embarrassed or distressed, because those records may be useful to investigators.
The government portal also allows citizens to report suspicious phone numbers, messaging handles, email addresses, SMS headers and websites even when an attempted fraud has not succeeded.
A contest between offensive and defensive AI
The banking system’s response will increasingly involve using AI against AI.
Financial institutions can deploy machine learning to detect unusual behaviour, identify networks of mule accounts, analyse malicious emails, prioritise security alerts and trace stolen funds. But attackers will continue experimenting with ways to evade those systems.
This creates an arms race in which models, attack patterns and defences evolve continuously.
India’s banks possess strong financial buffers, sophisticated payment infrastructure and an increasingly coordinated regulatory framework. Those advantages matter. But the RBI’s warning shows that capital adequacy alone cannot guarantee stability in a financial system where trust, identity and payment instructions can all be convincingly fabricated.
The next major test of banking resilience may not begin with a recession or a wave of loan defaults. It may begin with a phone call that sounds completely genuine.
“In the AI era, the central security question is no longer whether a message looks or sounds real. It is whether the request can be independently verified before money, data or access is surrendered.”
