AI Hackers Target Banks at Machine Speed, CERT-In Warns

The detail that makes this advisory different from the dozens of warnings CERT-In issues each year is the speed and the method. According to the agency, the attackers are using publicly available artificial intelligence tools to automate their break-ins "at machine speed". The threat, CERT-In said, is "active and ongoing". That is not the language of a historical incident report. It is the language of something still happening.

The advisory, reported by Reuters on 5 November 2026, lands at an awkward moment for Indian finance. The country's non-bank lenders and fintech firms have spent the last few years digitising almost everything — loan approvals, disbursements, customer onboarding, collections — and doing it fast, often with small security teams and heavy reliance on third-party software. Cryptocurrency exchanges, meanwhile, operate in a regulatory grey zone where the rules are still being written. All three are attractive targets for the same reason: they move money, and they move it through software.

What makes the Crafty Falcon warning unsettling is how little is known, and how much is implied. CERT-In did not name a single victim company. It did not say how much money was taken, or even whether money was taken in every case. It did not say where the group operates from, or who is behind it. What it did say was enough to put every financial firm in the country on notice: a group is out there, it is using AI to work faster than human defenders can respond, and it is already inside the sector.

What CERT-In actually said

It is worth being precise about the advisory, because the gaps in it matter as much as the contents. CERT-In said the group, which it calls Crafty Falcon, has breached non-bank lenders, fintech firms and cryptocurrency exchanges. It said the attackers use publicly available AI tools to automate break-ins "at machine speed", and it described the threat as "active and ongoing".

The agency added a detail that should worry any chief information security officer in financial services. The group, it said, "surveys its targets with machine intelligence" — in other words, it uses AI to scout its victims before it attacks them — and it moves money by invoking the victims' own payment and disbursement systems. That last point is important. The attackers are not necessarily breaking into a vault and carrying cash out the door. They are, according to CERT-In, getting the victim's own systems to do the moving for them: triggering legitimate payment rails, disbursement workflows and transfer mechanisms from the inside.

That approach is harder to detect than a crude smash-and-grab. A payment that travels through a company's own disbursement system looks, at first glance, like a payment the company made itself. Spotting it requires the kind of monitoring that many smaller lenders simply do not have: anomaly detection on transaction flows, tight controls on who — or what — can invoke a payment, and the ability to freeze a disbursement the moment something looks wrong.

CERT-In named four tools in its advisory: ARTEX, PentAGI, Cairn and CyberStrikeAI. Security researchers have linked at least two of them to China, according to Reuters. One of them, the China-made tool ARTEX, has been used in attacks on South Korean banks since late September, according to CrowdStrike. That detail connects the Indian advisory to a wider pattern of AI-assisted attacks across Asia, and it suggests Crafty Falcon — or the tooling it uses — is not a one-country problem.

But the advisory is also striking for what it does not say. CERT-In did not attribute the breaches to any country, government or organisation. It did not name any affected company. It did not give a figure for stolen funds, and it did not describe the scale of the damage. For a sector that runs on trust, that silence is both understandable and frustrating. Companies that have been breached are rarely eager to say so, and an agency that names victims without their consent would quickly find that nobody reports anything to it at all. The result is a warning that is simultaneously urgent and vague — and firms are left to fill in the blanks themselves.

How an AI-driven break-in works

To understand why CERT-In's language matters, it helps to picture what "at machine speed" actually means in practice. A traditional cyberattack is a human endeavour with human bottlenecks. Somebody has to find a target, probe its defences, find a weakness, write or adapt an exploit, get inside, look around, find the valuable data or the money, and get it out. Each of those steps takes time, and each of them leaves traces.

AI does not remove those steps. It compresses them. An AI agent can scan thousands of potential targets in the time it takes a human to examine one. It can test login pages, probe APIs, read error messages and adapt its approach in seconds. It can sift through stolen credentials, try them across dozens of services, and find the ones that still work. It can read a company's public documentation, map its systems from the outside, and build a picture of where the weaknesses are likely to be — all before a human attacker would have finished their first cup of coffee.

That is what CERT-In appears to be describing when it says the group "surveys its targets with machine intelligence". The reconnaissance phase of an attack — traditionally slow, manual and noisy — is being handed to software that does it quickly, quietly and at scale. And once the attackers are in, the same automation applies to the next phase: finding the payment systems, understanding how disbursements are triggered, and invoking them.

None of this requires a secret, military-grade AI system. CERT-In's point is precisely the opposite. The tools involved are publicly available. Some are open-source projects. Some are commercial products. The barrier to entry for this kind of attack has fallen sharply, and it will keep falling. A group that once needed a team of skilled operators can now do comparable work with a handful of people and a set of AI agents doing the heavy lifting.

That has uncomfortable implications for defence. Security teams have spent years building processes around human-speed attacks: alerts reviewed by analysts, incidents triaged in hours, patches deployed in days. An attack that moves at machine speed does not respect those timelines. If an AI agent can go from first probe to a fraudulent disbursement in minutes, then a defence that depends on a human noticing an alert and acting on it is already too slow. The answer is not to hire more analysts. It is to automate the defence as well — to build systems that can spot anomalous payment behaviour and stop it without waiting for a human to sign off.

The four tools CERT-In named

CERT-In named four tools, and the list shows how accessible this kind of attack has become. None is described as a secret or restricted system; in the agency's framing, they are publicly available. If Crafty Falcon can do this work with off-the-shelf AI, so can others.

Researchers have linked at least two of the four to China, according to Reuters, with ARTEX the clearest example. But that link is about the tooling, not necessarily about who directs Crafty Falcon. CERT-In did not attribute the group to any country, and that distinction matters: a tool made in one country can be used by an attacker anywhere else.

A wider pattern, not a single incident

Crafty Falcon is not the only case in which AI agents appear to have done the work of a hacking crew. On 4 November 2026, Acronis Threat Research Unit published a write-up it called "BOSSAI", attributing a campaign "with high confidence" to a single person in Hualien City, Taiwan. According to Acronis, that person pointed AI agents at hundreds of organisations in eleven countries.

Acronis says it recovered more than 90,000 files, about 9GB, from an open directory on the attacker's own server. Twelve agents were involved, nine of them active. Grok, OpenClaw, Claude Code and Codex built infrastructure, while Hermes, Cursor, Trae and CyberStrike ran the attacks. Acronis says it confirmed data theft or system access at about ten targets in three countries.

If accurate, the danger is not only that established groups are getting faster, but that small actors can now operate at a scale that once required an organisation. One caveat is essential, though. Acronis sells security products, and its findings have no independent reproduction, so the account should be read as a vendor's investigation, not settled fact.

The GitBook breach

The same week brought another reminder that AI-agent claims need careful handling. Documentation host GitBook was breached in what it called an AI-agent-driven intrusion, and its app was offline for about 16 hours. According to GitBook, the attacker exploited a flaw in single sign-on login to create sessions as GitBook staff — to log in as if they were employees.

That is a serious breach whatever tool was used. But the AI element remains unproven: GitBook named no model or tool and released no agent logs, so its AI claim is unconfirmed. It may well be true, but without logs or a named tool, outsiders cannot verify it. The case shows how quickly "AI-driven" can become the default description of an attack, even when the evidence has not been shared.

What financial firms should do now

There is no single fix for an attack that moves at machine speed, but the shape of a sensible response is clear: accept that human-speed defence is no longer enough. Firms should put anomaly detection on payment flows, so a disbursement that does not fit the usual pattern is flagged or stopped before the money leaves. They should tighten controls on who — or what — can invoke a payment, so a compromised account or stolen session cannot trigger a transfer on its own. They should be able to freeze a disbursement quickly when something looks wrong, and automate the defence so it can act in minutes or seconds, not hours.

None of this is exotic. It is basic financial hygiene, applied with the urgency the warning demands. Smaller lenders and fintech firms, with the thinnest security teams, will find it hardest — which is why the advisory matters most for them.

Conclusion

The Crafty Falcon advisory is short on names, figures and attribution, and that vagueness is frustrating. But its core message is clear: AI tools are being used to break into Indian financial firms and move money through their own payment systems, at a speed human defenders cannot match alone. The wider reports from the same week suggest a pattern, not an isolated incident, even if each claim must be weighed on its own evidence. The response is not panic: build defences that work at the same speed as the attacks.