CERT-In warns of CRAFTY FALCON AI financial attacks siphoning payment APIs

Key facts

Advisory escalation and financial attack surface

On Wednesday 4 November 2026, the Computer Security Incident Response Team for Finance (CSIRT-Fin) under India's national cyber agency, CERT-In, issued Critical Advisory CIAD-2026-0058 warning financial institutions of an escalating campaign orchestrated by an AI-assisted threat actor designated CRAFTY FALCON. The advisory, updated on 6 November 2026, marks an acute escalation from an earlier alert issued on 9 October 2026 (CIAD-2026-0046), reflecting 26 days of continuous adversarial adaptation across the Indian financial ecosystem (derived).

CRAFTY FALCON combines autonomous multi-agent frameworks, machine intelligence reconnaissance, and automated script generation to compromise six distinct financial sub-sectors (derived): non-banking financial companies (NBFCs), digital lending platforms, fintech startups, payment aggregators, payment service providers, and cryptocurrency exchanges. Operating at machine speed, the threat actor targets internet-exposed test environments, reverse-engineers application code, modifies mobile and web application packages, and executes unauthorized payment and disbursement API calls. To confirm payment routes before executing catastrophic transfers, the actor conducts micro-value test transactions of one rupee (Rs. 1, approximately $0.012 USD, derived). CERT-In emphasized that CRAFTY FALCON deliberately attacks during weekends, holidays, and off-peak monitoring windows, systematically re-targeting entities that performed incomplete remediation.

Zoom

Advisory timeline and operational progression of CRAFTY FALCON campaign

Date	Agency Event	Operational Milestone	Analytical Significance
9 October 2026	Initial CERT-In Alert	CIAD-2026-0046 issued under TLP:AMBER to select BFSI entities	First baseline warning on autonomous reconnaissance scripts
4 November 2026	Critical Escalation	CIAD-2026-0058 published detailing live payment API siphoning	26 days elapsed; confirmation of multi-agent toolchains (derived)
6 November 2026	Public Sector Update	Comprehensive remediation directive covering 6 BFSI sub-sectors	Mandatory replacement of IP allow-lists with mutual TLS (derived)

A alarming finding in CIAD-2026-0058 is the adversary's deliberate operational schedule. CRAFTY FALCON times its intrusions to coincide with weekends, national holidays, and overnight coverage gaps when security operations center (SOC) staffing is reduced and alert thresholds are frequently relaxed. Furthermore, CSIRT-Fin confirmed that several victim organizations that previously received private TLP:AMBER alerts following the 9 October advisory were successfully breached a second time. Attackers exploited lingering persistence mechanisms, unrotated tokens, and unpatched staging systems left behind after hasty, incomplete initial cleanups.

Tactical arsenal: four autonomous AI frameworks

According to forensic analysis conducted by CSIRT-Fin, CRAFTY FALCON executes complete intrusion lifecycles autonomously by chaining four specialized, publicly accessible artificial intelligence offensive frameworks. Rather than deploying static precompiled malware, CRAFTY FALCON uses generative models to synthesize large volumes of tailored, functionally identical scripts in rapid succession. This continuous code mutation defeats traditional static hash signatures and behavioral heuristics.

Zoom

Chained offensive AI toolkits identified in CRAFTY FALCON campaigns

Offensive Toolkit	Architectural Role	Primary Capability	Adversarial Advantage
ARTEX LLM Family	Code Analysis & Synthesis	Decompiles client applications and audits disbursement logic	Generates API call harnesses in seconds
PentAGI Framework	Autonomous Orchestration	Multi-agent planner coordinating dynamic lateral movement	Adapts execution paths without human operators
Cairn Recon Engine	Asset Discovery	Scans public ranges for exposed staging and UAT endpoints	Discovers uncataloged, forgotten devops subdomains
CyberStrikeAI	Polymorphic Mutation	Feedback-driven exploit script generation and evasion	Defeats static SHA-256 hashes and standard EDR

The integration of PentAGI represents a qualitative shift in adversary capability. PentAGI operates as a coordinator agent that delegates sub-tasks to specialized worker agents, analyzing network topology responses in real time. If a staging boundary blocks direct database queries, the agent re-routes queries through intermediate test microservices or extracts credentials stored in client environment variables.

Staging ingress, IP bypass, and micro-transaction validation

CERT-In documented a multi-stage intrusion methodology that systematically exploits procedural oversights in modern cloud development. Attackers begin by targeting internet-accessible User Acceptance Testing (UAT), staging, and quality assurance environments. These environments frequently mirror production application logic and database schemas but lack enterprise-grade monitoring, web application firewalls, and multi-factor authentication.

Upon securing a foothold in non-production environments, the actor extracts hardcoded API tokens, database connection strings, and shared cloud credentials to transition laterally into live production clusters. CRAFTY FALCON then reverse-engineers client applications and mobile software packages. Threat actors alter the application code to initiate unauthorized payment requests directly against core disbursement APIs. Because defenses assume API traffic from valid application processes is legitimate, the altered packages bypass standard fraud tripwires.

Financial institutions frequently protect internal disbursement APIs by restricting access to known partner IP ranges. CRAFTY FALCON bypasses these restrictions by routing requests through shared third-party fetcher IP pools, including content delivery networks, application store review scanners, and URL preview services, which can be triggered by external actors to make outbound web requests to target APIs. In AWS environments, the threat actor abuses legitimate administrative management services, specifically AWS Systems Manager (SSM) and AWS Lambda, blending operational activity into normal administrative audit streams.

Before draining organizational reserves, CRAFTY FALCON verifies transaction plumbing using minimal test values. Attackers submit test transactions valued at Rs. 1 (approximately $0.012 USD, derived) to newly added beneficiary accounts. Once automated systems confirm that settlement rails process the micro-transaction without generating alerts, the group initiates high-velocity transfers through merchant settlement APIs and payment aggregators. In campaigns targeting cryptocurrency exchanges and virtual digital asset (VDA) providers, the threat group locates and exfiltrates private wallet keys from cloud storage, transferring crypto assets directly into unhosted attacker-controlled wallets.

CERT-In mandatory mitigation directives

In response to the escalating threat, CERT-In and CSIRT-Fin published immediate operational directives for all financial entities. Foremost among these is the complete isolation of non-production infrastructure from the public internet. Organizations must eliminate all internet routes into UAT and testing subdomains and terminate any shared network routes or credentials connecting test clusters to production.

Zoom

Mandated technical controls and verification procedures for financial platforms

Control Domain	Technical Requirement	Targeted Attack Vector	Verification Standard
Environment Isolation	Air-gap all UAT / staging infrastructure from the public internet	Cairn recon and staging credential harvesting	Zero external DNS resolution for test subdomains
API Authentication	Mandate mutual TLS (mTLS) and deprecate partner IP allow-lists	Third-party fetcher IP pool proxy evasion	Client certificate validation on all settlement routes
Fraud Tripwires	Flag Rs. 1 validation transactions followed by burst transfers	Micro-value settlement rail verification	Mandatory cooling window on newly added beneficiaries
Cloud Governance	Lock down AWS Systems Manager (SSM) and Lambda execution roles	Administrative control plane subversion	Real-time alerts on off-hours script execution tasks
SOC Operations	Maintain full analyst staffing and lowered thresholds on weekends	Adversarial off-hours timing and re-targeting	Continuous 24x7 monitoring with zero off-peak blindspots

CERT-In emphasized that organizations recovering from suspected compromises must perform exhaustive forensic sweeps rather than localized system reboots. Given CRAFTY FALCON's pattern of re-targeting partially cleaned networks, security teams must rotate all database connection secrets, revoke master cloud IAM tokens, audit Lambda execution bundles, and verify that no auxiliary persistence tasks remain active across AWS Systems Manager agent configurations.