AI security
Crafty Falcon: AI-driven attacks on payment APIs
CERT-In's Critical advisory CIAD-2026-0058 names Crafty Falcon, an AI-assisted actor abusing fintechs' own payment and payout APIs to move money fast.
India's Computer Emergency Response Team (CERT-In) has put a Critical rating on something unusual: an attacker that moves money without breaking anything. Its advisory CIAD-2026-0058, issued on November 4, 2026, names a threat actor it calls Crafty Falcon and says the group is draining funds from financial firms by calling those firms' own payment and disbursement application programming interfaces (APIs), the software interfaces a business uses to send payouts and settle transactions. The actor surveys each target with what CERT-In calls machine intelligence, at machine speed, then invokes those interfaces the way a legitimate system would. There is no exploit to catch and no malware signature to match.
The way in is the part most teams will recognise with a wince. CERT-In says Crafty Falcon enters through internet-exposed test, user-acceptance-testing (UAT) and staging environments, the copies of production that carry real credentials and loose controls because nobody expects them to face attackers. From there it harvests shared credentials, tokens and network paths that reach the live systems, and it also targets the victim's cloud infrastructure, including compute, serverless and management services. The advisory covers non-banking financial companies (NBFCs), which are lenders that are not banks, along with digital lending platforms, fintechs, payment aggregators, payment service providers and cryptocurrency exchanges. It updates an earlier advisory from October 9, and CERT-In treats the campaign as active and escalating.
How it works
A payment API lets one system tell another to move money. A lending app calls a disbursement API to pay out a loan, a marketplace calls one to settle with a seller. The interface trusts whoever holds a valid credential and sends a well-formed request. Crafty Falcon does precisely that. It needs no memory-corruption bug and no zero-day, because the API is doing its job: it received an authenticated instruction to disburse funds and it disbursed them. That is why CERT-In frames this as abuse of legitimate functionality, not exploitation of a flaw.
Reports on the advisory describe the actor probing a disbursement route with a token-sized transfer first, in some cases a single rupee, to confirm the path works before pushing high-value transactions through it. Because each call looks like normal business traffic, the tell is not the individual request but the pattern: a newly added payee followed by a burst of payouts, or a run of transactions faster than any human operator could submit.
The test and staging systems are the door into production. A staging copy of a payment platform usually holds working credentials so integration tests pass, and it often sits on the internet so partners can test against it. CERT-In says the group collects those shared credentials, tokens and network routes and reuses them against the real systems, so an environment built for convenience becomes the shortest path to the production payout engine. A test environment that answers from the public internet is one of the first things we flag in a web and API penetration test.
Crafty Falcon also goes after the cloud control layer. The advisory lists compute, serverless and management services as targets, which means the actor is not only calling payment APIs but reaching the cloud accounts and management consoles that run them. Control of the management plane lets an attacker change configuration, read secrets and create its own access, a far more durable foothold than any single stolen key.
For cryptocurrency platforms the goal is more direct. CERT-In says that where the target is an exchange or virtual digital asset provider, the group steals wallet keys and empties the wallets, since whoever holds the private key controls the funds with no API in the loop at all. Tying all of this together is the reconnaissance: CERT-In says the actor uses machine intelligence to map exposed environments, credentials and API routes far faster and more thoroughly than a human team scanning by hand, and it reportedly leans on publicly available offensive security tooling to do so.
What CERT-In is warning
CIAD-2026-0058 is an update to a restricted advisory CERT-In issued on October 9, written by CSIRT-Fin, the agency's financial-sector incident team. The November version raises the alarm because the problem got worse. CERT-In says it is still receiving reports of successful attacks, including from firms that had received the earlier advisory and already carried out remediation. Full indicators of compromise were shared only with financial institutions in a restricted copy, so the public advisory describes behaviour rather than specific addresses or hashes.
The timing is deliberate. CERT-In says attacks have been aligned with holidays, weekends and other low-monitoring windows to delay detection and response, and that the same entities have been re-targeted, including after incomplete remediation. A fintech that closed one exposed test server over a long weekend, and missed the next one, is exactly the profile the advisory describes.
One caveat matters for how you read this. CERT-In's advisory does not establish that all of these incidents were necessarily carried out by a single actor, and it has not named affected firms or disclosed how much has been stolen. Crafty Falcon is a label for a pattern of AI-assisted financial intrusions, not proof that one group is behind every reported loss.
What to do
- Separate test, UAT and staging from production, and starve them of anything real. They must not share credentials, tokens, API keys or network paths with the live environment, and they should not hold live payment keys or production data. Put them behind authentication and off the public internet; a staging payment platform does not need to be reachable from the world.
- Put real controls on the payment and disbursement APIs. Require strong, per-client authentication, enforce strict rate limits so one credential cannot fire hundreds of payouts a minute, and add anomaly detection that alerts on off-hours bulk disbursements and on new payees. CERT-In recommends velocity rules and a cooling period before a newly added beneficiary can receive high-value transfers, which breaks the test-then-drain sequence.
- Treat the cloud management plane as crown-jewel access. Cut cloud and backend roles to the least privilege each workload actually uses, apply default-deny egress so a compromised service cannot call out freely, and require phishing-resistant multi-factor authentication (MFA), such as hardware security keys, on every account that can change configuration or read secrets.
- If you run a crypto platform, keep wallet keys out of reach. Hold private keys in a hardware security module (HSM), a dedicated device that performs signing without exposing the key, or in multi-party or cold custody rather than in application config. Require multi-signature approval and withdrawal limits for payouts, so no single stolen key empties a wallet.
- Staff or automate monitoring across weekends and holidays. The advisory calls for round-the-clock transaction surveillance, and the attacks are timed for the hours when staffing is lowest. If you cannot run a 24-hour desk, make the response automatic: have off-hours bulk-payout alerts page someone, or freeze payouts until a human approves them.
- Report suspicious activity to CERT-In and to the Indian Cyber Crime Coordination Centre (I4C). If you received the October advisory and remediated, check again. CERT-In says attackers have come back for firms that fixed only part of the problem, so confirm every exposed environment is closed, not only the one that was hit.
The gap Crafty Falcon walks through is an exposure and API security problem before it is a fraud problem: test systems reachable from the internet, payment endpoints that trust any valid token, and a management plane one stolen credential away. For crypto platforms, our smart contract and wallet audits cover key custody and withdrawal controls. Our security operations work covers the round-the-clock monitoring the advisory asks for, and our cloud and Kubernetes security reviews check role permissions and management-plane access. Open the chat and Yaali, our AI agent, will pass your question to the engineer who would do the work.
Sources: CERT-In advisory CIAD-2026-0058, Infosecurity Magazine, Moneylife, YourStory, Outlook Money, The Register, LetsDataScience.