The Small Bot Can Be the Bigger Risk
America's bank regulator is telling examiners to judge AI by its authority and the consequences of failure, not by how impressive the model sounds.
Read the pick: The Small Bot Can Be the Bigger RiskX OPPORTUNITY RADAR / QUESTIONS WORTH ASKING
CERT-In now calls the payment-API campaign active and ongoing, says successful repeat attacks followed incomplete remediation, and documents AI-assisted tooling that turns ordinary exposed systems and credentials into machine-speed theft.

Proof, not another handsome policy binder. Reuters reports that the Reserve Bank of India has been pressing lenders in recent supervisory meetings for evidence of compliance with existing cybersecurity rules, testing of API weaknesses and preparedness for AI-related risks.[1]
That account rests on three unnamed sources familiar with private discussions. Reuters says the RBI did not immediately respond to its request for comment. Treat the supervisory push as attributed reporting—not a public RBI announcement or a newly issued rule.[1]
Reuters reports that two Mumbai-based non-bank lenders lost about $2 million through more than 700 unauthorized instant transfers on October 8–9. A police complaint reviewed by Reuters described a compromised API used to send loan-disbursement instructions to ICICI Bank. The bank did not immediately comment, Reuters said.[1]
The report also describes a separate leak involving roughly 3,900 State Bank of India customer records. SBI said it was investigating; the exploited vulnerability was not known. Do not merge those incidents into one proven attack chain.[1]
An approved connection is useful plumbing. It is not a character reference for whoever happens to be holding the credentials, droogies.
CERT-In’s October 9 critical advisory independently describes attacks against NBFCs, digital lenders, fintechs, payment providers and wallets. Its documented pattern runs from compromised exposed applications, APIs or privileged endpoints to stolen payment or partner-bank API credentials, then direct fund-transfer calls from the victim’s whitelisted infrastructure.[2]
The agency warns that this can bypass application-level checks on KYC, loan approval, beneficiaries, balances and transaction limits. It describes dummy transaction references, mule accounts and rapid transfers. A trusted route can carry an untrusted instruction.[2]
CERT-In also discusses possible AI-assisted offensive tooling. That is not proof that AI caused either named incident. Nor does its advisory independently confirm Reuters’ account of the private RBI meetings.[1][2]
The advisory calls for an API inventory, retirement of unused endpoints, stronger authentication, credential controls, validation tied to transaction stages, maker-checker approval, payout limits and monitoring. The common point is to verify the transaction—not merely recognize the machine sending it.[2]
The regulator reportedly wants evidence that controls survive contact with the network. The technical warning already gives institutions something less glamorous and more useful than a slogan: examine where trusted connectivity is being mistaken for permission to move money.
CERT-In now rates the escalation Critical and says CSIRT-Fin has named the active campaign CRAFTY FALCON. The agency says it continues to receive reports of successful attacks, including against financial institutions that had already received restricted advisories and attempted remediation.[3][4]
Some institutions were deliberately attacked again after incomplete repairs, CERT-In says, with operations timed for holidays, weekends and other low-monitoring windows. The advisory describes publicly available AI-assisted offensive tools that run parallel attack chains, rapidly generate tailored scripts and adapt payloads to target responses.[3]
The futuristic label should not hide the ordinary entry points. CERT-In identifies exposed testing and staging systems, shared credentials and network paths into production, permissive cloud roles, reachable old APIs and plaintext payment keys. The AI tooling compresses the time between discovery and theft; it does not invent the unlocked door.[3]
The public advisory does not identify every victim or quantify total campaign losses. It establishes an active threat pattern and official mitigations, not proof that every recent Indian financial breach used the named tooling or followed the same sequence.[3][4]
Update ·
Added CERT-In’s November 4 Critical escalation notice: the agency named the active CRAFTY FALCON campaign, reported successful repeat attacks after incomplete remediation and documented the conventional exposed systems and credentials that the AI-assisted tooling exploits.
Checked Nov 4, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.