Indian cybersecurity agencies say criminals are using AI agents to move money out of banks at machine speed. Other regulators are tightening their scrutiny, but the number of victims and losses has not been published.
India's cybercrime coordination centre has warned banks, non-bank lenders, and other financial firms that criminals are using artificial intelligence agents to carry out financially motivated attacks [1]. The advisory was made public on November 4, 2026, and it asks institutions to apply measures that CERT-In, the national cybersecurity agency, set out on October 9 [1].
The warnings are coming from more than one place. Regulators in the United States and Europe are reportedly asking harder questions about AI risk, including autonomous agents and new cyber threats [4]. What follows covers what has been published, what it says, and what remains unknown.
What India's advisory says about the attacks
The advisory describes attacks that move at the speed of software rather than people. According to Reuters, the agency calls the campaign "machine speed" compromise, carried out with agentic AI frameworks that can run a whole intrusion without a human at each step [1].
The targets are not limited to big banks. The advisory names non-bank lenders, digital lending platforms, fintech firms, payment aggregators, payment service providers, and cryptocurrency exchanges [2]. Some institutions that had already received earlier advisories and started remediation were hit again, which the agency attributes to fixes that were incomplete [2].
Timing is part of the method. The attacks reportedly cluster around holidays, weekends, and periods of reduced monitoring, when a delayed response gives the attackers more room [2].
How the money leaves: through the bank's own payment interfaces
The most useful detail in the advisory is how the funds are taken. According to CERT-In's description, the actor steals money by invoking the victims' own payment and disbursement APIs [2].
That changes where you look for the problem. If the attacker is calling interfaces the institution already exposes, the question is less "how did they get past the login" and more "what can one authorized-looking request do once it arrives?" Stolen funds are reportedly sent to mule accounts at other banks and withdrawn quickly through IMPS or NEFT, and the actors then try to erase traces [3].
The attacker name is a regulator's judgment
CERT-In designated the AI-assisted actor "CRAFTY FALCON" and attributes its activity to publicly available AI-driven offensive security frameworks [2]. The advisory names several of those tools. I have left them out because defenders gain nothing from the list, and I have not independently confirmed the attribution.
Treat a naming like this as a regulator's assessment. It can guide where to look, but it is not a court finding, and it does not say how many institutions were hit.
Six hours is the new clock
Reporting rules are tightening alongside the warnings. Institutions are directed to report incidents to CERT-In within six hours of detection, and banks and NBFCs must also notify the Reserve Bank of India through its DAKSH portal within the same window [3].
Six hours is short when a fraud can move money in minutes. The practical consequence is that the decision-making has to be done before the incident. Who can freeze payment APIs? Who can approve a regulator report at 2 a.m. on a holiday? A tested incident response plan answers those questions in advance.
What other regulators have said
In the United States and Europe, banking regulators are reportedly asking about autonomous AI agents, foreign-developed models, and new cyber risks, according to people familiar with their questions [4]. Those reports come from unnamed sources, so treat the detail as reported rather than confirmed.
The Bank of England and the IMF have also warned that AI-powered cyberattacks and deepfake scams are eroding trust in the financial system [5]. The OECD's incident monitor classed that warning as an AI hazard rather than an incident, because it described plausible harm and did not detail a specific case [5]. That distinction matters. These are warnings about what could happen, not a documented wave of losses in those countries.
| Date | Body | What was reported | Status |
|---|---|---|---|
| Oct 9, 2026 | CERT-In | Earlier measures for financial institutions | Reported, per Reuters [1] |
| Oct 13, 2026 | US and European banking regulators | Stepped-up scrutiny of AI risks | Reported, unnamed sources [4] |
| Oct 14, 2026 | Bank of England and IMF | Warnings on AI cyber fraud and deepfakes | Reported via OECD monitor [5] |
| Nov 4, 2026 | I4C | Advisory on machine-speed fraud by AI agents | Public advisory, per Reuters [1] |
| Nov 4, 2026 | CERT-In and CSIRT-Fin | Critical advisory CIAD-2026-0058 on the BFSI campaign | Republished copy [2] |
What has not been established: the number of victims, total losses, whether the same actor is behind attacks in different countries, and which AI tools were used in any particular case. Reports of multi-crore transactions describe individual cases, not a total.
What to check this week
If the advisory's description holds, the first place to look is the payment and disbursement APIs. Check who can call them, what a single call can move, and whether one credential can trigger bulk transfers. Per-call and per-day limits are cheap to set and hard for an attacker to work around quietly.
Next, look at beneficiaries. Payments to newly added payees, sudden bulk disbursements, and transfers into recently opened accounts should trigger a hold or a human review. Mule accounts are the step where the money becomes hard to recover, so that's where delays help most.
Then check coverage. The advisory's advice on round-the-clock monitoring, staffing over holidays and weekends, and lower alert thresholds at quiet times is practical, and many teams run thinner at exactly those times. Our piece on autonomous AI attacks covers why the defender's clock is getting harder to keep up with.
Regulators also flag deepfakes, so voice and video confirmation for high-value requests belongs in the same review. The mechanics of AI-enabled vishing and deepfake social engineering are a separate problem, but the controls overlap.
What to watch next
The next useful checkpoints are the full CERT-In advisory text, any RBI circular that follows the DAKSH reporting rule, and whether other national regulators publish comparable advisories. The Bank of England and IMF primary documents are also worth reading in full before anyone quotes their wording.
Sources & References
- Reuters (syndicated), "India cybercrime agency warns financial firms of AI-driven cyberattacks" (November 4, 2026)
- Ministry of Cyber Affairs, republished CERT-In advisory coverage, "Machine-Speed Financial Theft" (November 4, 2026)
- Deccan Herald, "Cyber agency warns banks, fintech firms of AI-powered frauds involving multi-crore transactions"
- Cyprus Mail (Reuters-based), "Banking regulators step up scrutiny of AI risks" (October 13, 2026)
- OECD.AI incident monitor, "AI-Driven Cyber Fraud Threatens Trust in Financial System, Warn BoE and IMF" (October 14, 2026)
Leave a Comment (Authenticated Users)