I4C says AI agents are emptying payment APIs at machine speed. Banks have six hours to call CERT-In.

The problem in plain words

On 3 November 2026, the National Cybercrime Threat Analysis Unit under the Indian Cyber Crime Coordination Centre issued an advisory that Inc42, The Economic Times, PTI and The Hindu BusinessLine all reported the next day. Attackers are no longer only phishing one customer at a time. They are using artificial intelligence agents to find weak spots in bank and fintech systems, take over the automated pipes that move money, and fire off transfers before a human security team notices. I4C said individual cases already show multi-crore unauthorised transactions through fund-disbursement systems and payment APIs.

An API is simply a software doorway that lets one system ask another to do something, such as credit a beneficiary or push an IMPS payout. Banks and fintechs rely on dozens of these doorways for UPI, payroll, lending disbursals and merchant settlements. If an attacker controls the doorway, the theft looks like a normal payment. That is why this warning matters to anyone who uses Indian digital payments: the target is the plumbing behind the QR code, not only the customer's phone.

How the attack builds tension

According to the advisory, as summarised by Inc42 and Hindu BusinessLine, criminals first exploit unpatched software, misconfigured internet-facing servers and weak customer-facing or complaint-handling interfaces. Once inside, they aim at automated fund-disbursement systems and specialised payment APIs. I4C's phrasing is blunt: use of an agentic AI framework makes the attack possible at machine speed. Money then moves into mule accounts across banks and is withdrawn through IMPS or NEFT. Records are deleted so investigators lose the trail.

The Economic Times noted the warning lands days after the Reserve Bank of India issued a related alert. Neither I4C nor the news reports named specific banks. That silence is itself a stake: customers cannot tell which pipes were hit, only that the pattern is real enough for the Home Ministry's cybercrime centre and MeitY's CERT-In to speak together.

That combination of speed and silence is what raises the danger. Older cyber frauds often needed a victim to click a link or read an OTP aloud. An agentic API compromise can empty an automated disbursement queue while the victim's phone stays quiet, which is why I4C's emphasis on inventorying every payment interface and disabling leftover test endpoints is more than housekeeping. Each forgotten doorway is another place a machine can knock.

What defenders are told to do

I4C and CERT-In told financial institutions to implement critical-severity CERT-In advisories on payment and disbursement APIs and on AI agents without delay, and to align with CERT-In's May 2026 blueprint against AI-assisted vulnerability exploitation. Inc42 listed practical steps: keep an inventory of payment-related APIs, disable obsolete and test interfaces, set transaction limits, verify beneficiaries, watch sudden jumps in speed or volume, finish KYC and approval checks before funds move, and run authorised red-team drills that simulate machine-speed attacks.

The clock is short. Institutions must report incidents to CERT-In within six hours of detection. Banks and NBFCs must also report to the Reserve Bank of India through the DAKSH portal in the same window. That dual reporting is meant to freeze mule flows before the money leaves the domestic network.

The wider fraud backdrop

Hindu BusinessLine relayed Union Home Ministry figures from I4C's National Cybercrime Reporting Portal and Citizen Financial Cyber Fraud Reporting system: more than 53.87 lakh cyber fraud complaints between financial years 2023-24 and 2025-26, more than 1.81 lakh related FIRs, reported amounts above Rs 56,087 crore, liens above Rs 9,079 crore, and refunds to victims of a little over Rs 206 crore. Those numbers show why a machine-speed API attack is so dangerous. Even when humans eventually notice, most stolen value never comes back.

What could still go wrong

Best case is that banks and fintechs patch the critical CERT-In items this week, shut unused APIs, and catch the next agentic run inside the six-hour window. Worst case is that mule networks stay faster than compliance teams, test interfaces stay online for convenience, and customers only learn after a UPI or loan disbursement pipe has already emptied. For builders in India's digital economy, the actionable split is clear: treat every payment API as a high-value vault door, assume AI agents will probe it continuously, and design monitoring that reacts at machine speed rather than after the morning stand-up. For ordinary account holders, the practical habits still matter: lock biometrics where the bank allows it, treat unexpected video-KYC or recovery calls as hostile until verified through the official app, and report suspected fraud immediately on cybercrime.gov.in so liens can be placed before IMPS and NEFT mop up the trail.