Could Your Financial Institution Stop a Crafty Falcon Attack?
What banks and NBFCs should do now to detect, contain and recover from abuse of payment and disbursement APIs.
Perspectives · Trust by Design · Operational Resilience
A low-value payout goes through during a holiday shift. The credentials are valid and the request comes from a familiar network. Later, larger payouts follow. If the institution checks only the credential, source IP and per-transaction limit, each request can look routine while the pattern is anything but.
On 4 November 2026, CERT-In issued CIAD-2026-0058, an update on the AI-assisted campaign CSIRT-Fin calls Crafty Falcon. The advisory describes abuse of financial institutions’ own payment and disbursement APIs, including merchant settlement interfaces. It reports exposed credentials, misuse of trust in shared network addresses, small test transfers before larger ones, activity during low-coverage periods and repeat targeting after incomplete fixes.
The advisory describes AI assistance to the attackers. It does not establish that a bank’s own AI system had to be compromised. A financial institution can have no AI in its payment flow and still be exposed. The immediate test is whether the institution can detect and stop a technically valid request that has no legitimate business authority behind it.
If you suspect unauthorised payouts, act now
Activate incident command and bring security operations, payments operations, treasury, fraud, technology, and compliance into the same incident. Give the incident lead a named payments counterpart with authority to contain affected flows. Identify which payment routes, credentials, beneficiaries and settlement batches are in scope. Where the evidence warrants it, revoke the affected credential, isolate the compromised host, or pause the specific outbound route. Work with the relevant bank, processor or payment rail to trace transfers and pursue any holds or recalls still available.
Preserve gateway, identity, endpoint, build and payment records with reliable timestamps. Correlate the request with the business instruction, ledger entry, settlement record and any reversal. An API success response, ledger posting and final settlement are different facts. Initiate reporting under the institution’s applicable obligations and use the CERT-In and CSIRT-Fin channels listed in the advisory. Start that process while investigation is underway.
Before restoring service, verify the original entry point is closed, affected credentials are invalid wherever they were accepted, and alternate or recovery routes are clean. Review any held or uncertain payment against the ledger and the rail before release or replay. A reset without that reconciliation can turn recovery into a second payment incident.
Even without an alert, run these checks today
| When | Owner and action | What to show |
|---|---|---|
| First hours | Payments and platform owners list every route that can move money, including partner callbacks, older versions, batch jobs and recovery endpoints. | A current list of routes, credentials, limits and owners, including who can pause each route outside office hours. |
| Today | Security and payments operations review recent activity together. Look for a new beneficiary, small probes followed by larger payouts, unusual credential use and activity outside normal processing windows. | A trace from API call to settlement and evidence that an after-hours alert reaches someone who can act. |
| Next 24–72 hours | Engineering and fraud teams test credential revocation, blocked replays, changed payment details and recovery-route parity. Prioritise fixes on routes with the broadest payout authority. | Test results, named remediation owners and a scheduled retest. These are operational targets, separate from reporting deadlines. |
Make the payment service check the business instruction
A network allowlist can narrow access, but a shared vendor IP is a poor substitute for a verified caller. Give each integration a scoped identity. Make the service verify that the caller can execute this specific instruction and that its beneficiary, amount, currency and funding account match the authorised record. Where signed requests are used, cover the fields that matter, enforce a short validity window, and reject reused nonces within that window. Use payment-level duplicate controls as well: an attacker or a buggy client can send a fresh, correctly signed request for the same instruction.
Scan for secrets in repositories, build outputs and logs. Restrict outbound access from services that fetch URLs, and prevent those services from reaching sensitive internal endpoints. Watch for patterns across credentials, beneficiaries and routes; individual amounts can sit below every threshold. A signal such as a new beneficiary or a low-value payout warrants context and review. On its own it does not prove fraud or identify Crafty Falcon.
Ask for a demonstration before the next holiday
Can the overnight team identify which credential initiated an unusual payout, stop that credential without blocking unrelated payments, and tell treasury whether funds actually settled? Run the exercise on a primary route and a recovery route. The result the board should ask for is how long it took to detect, contain and account for the transaction, with a named owner for each gap.
Continue with Learning
The Saturday settlement alert. Practise triage and containment during an after-hours settlement alert.
Put this perspective into practice.
Explore the concepts, make a decision and test what changes when the situation changes.
Topics: API Architecture, Operational Resilience, Perspective, Trust by Design