On November 4, 2026, India's Computer Emergency Response Team published advisory CIAD-2026-0058, an update to a warning first issued on October 9. It names an AI-assisted actor, CRAFTY FALCON, and says the campaign against banks, NBFCs, fintechs, payment aggregators and crypto exchanges is active and escalating. CERT-In reports that attackers operate at machine speed and strike by calling the victim's own payment and disbursement APIs. The advisory names the AI tools involved, including ARTEX, PentAGI, Cairn and CyberStrikeAI. No exchange was named and no losses were quantified. Coverage includes Cointelegraph, BW Businessworld and Moneycontrol.

The path into production is not exotic. Attackers get in through vulnerable applications and internet-facing UAT, staging and test environments. They reuse shared credentials, access tokens and network links that connect non-production systems to production. They reverse-engineer client apps to craft unauthorised API calls, and they abuse payment aggregators' merchant settlement APIs. Before a large transfer, they often send tiny test payments, sometimes as small as one rupee. They prefer weekends and holidays, when staffing is thin, and they re-target organisations that patched incompletely. For crypto exchanges and other VASPs, CERT-In says wallet keys have been stolen and funds drained.

If you run an exchange, a VASP, a payment stack or any service that can move customer funds, the fix starts with two habits: isolate staging, and kill shared keys.

Isolate staging from production

Treat every non-production environment as hostile until proven otherwise. Staging, UAT and test systems should not sit on the open internet with the same trust as production, and they should not share network paths that let a foothold there become a foothold in prod.

Put staging in its own account, VPC or project. Block lateral movement with firewalls, private networking and allowlists. Do not reuse production databases, wallet services, HSM paths or payment endpoints from a test box. If developers need production-like data, use scrubbed fixtures, not live keys and not a tunnel that can call disbursement APIs.

When a test environment must talk to a real provider, give it a dedicated sandbox credential with no path to settle, withdraw or sign. If that credential is stolen, the blast radius should stop at fake money.

Kill shared credentials and API keys

Shared passwords, shared API tokens and long-lived keys that work in both staging and production are how a test compromise becomes a production drain. Every environment needs its own identities. Every human, service and CI job should have its own credential, scoped to the least privilege it needs, with rotation and revocation that you actually practise.

Wallet signing keys, withdrawal API keys and settlement credentials never belong in staging config, shared chat, or a developer laptop that also holds production access. Prefer short-lived tokens, hardware-backed or HSM-backed signing, and dual control for anything that can move funds. When someone leaves, or a vendor changes, revoke first and ask questions second.

If you discover that a key was ever shared across environments, assume it is burned. Rotate it, review every call made with it, and check whether staging still has a network path into prod.

Watch for the probe before the drain

CERT-In describes a clear pattern: a tiny unauthorised transfer first, then a large one. Build transaction monitoring that treats a small unexpected payment, a new beneficiary, or a first-time API path as a high-signal event, not noise.

Add velocity limits and cooling-off periods for new beneficiaries and for sudden spikes in disbursement volume. Alert on settlement API calls outside normal hours, from unexpected IPs, or using credentials that should only exist in non-production. For crypto specifically, watch for wallet export, key unwrap, signing spikes and withdrawals that skip your normal approval path.

Monitor cloud changes around the clock

Attackers in this campaign also target cloud infrastructure to deploy malicious files. Turn on real-time audit-log monitoring for remote command execution, unexpected role assumptions, and unauthorised changes to functions, containers, secrets and IAM policies. Staff monitoring through weekends and holidays, and lower alert thresholds in low-coverage windows.

What good looks like

Staging cannot reach production wallets or payment APIs. No credential works in more than one environment. Every fund-moving key has an owner, an expiry and an alert on use. Tiny test transfers against production paths page a human immediately. Incomplete remediation from an earlier incident has been closed, not left as a second chance for the same actor.

These are fundamentals, not advanced tradecraft. AI-assisted attackers compress the time between a foothold and a drain, but they still need the same weak doors: exposed staging and shared keys. Close those doors first.

For teams that want to go deeper on crypto operational security beyond this checklist, the Advanced Security Academy is coming. Join the waitlist on advancedsecurity.club.