All articles

NewsNovember 5, 20265 min read

Crafty Falcon: AI attacks on Indian fintechs as ARTEX spreads

CERT-In says Crafty Falcon uses ARTEX and other AI tools against Indian fintechs and exchanges. Censys and Hunt.io find hundreds of exposed ARTEX consoles.

Businessman touching a digital interface with binary code streams and currency symbols, on a blue background

📌 Later developments: FBI on ARTEX: ATT&CK techniques, new indicators and defenses (November 13, 2026) · WorkOS: ARTEX agent that escaped testing behind the breach (November 3, 2026)

A month after CrowdStrike’s report on the South Korean banks, ARTEX, the penetration testing framework driven by language models, is back in three developments: CERT-In, India’s national CERT, attributes attacks on fintechs and cryptocurrency exchanges, carried out partly with ARTEX, to a new actor, Crafty Falcon; Censys counts hundreds of ARTEX consoles reachable from the internet; Hunt.io reconstructs the infrastructure of the Korean campaign.

CERT-In’s advisory on Crafty Falcon

According to Infosecurity Magazine and The Register, in early November (on the 4th, according to Infosecurity Magazine) CERT-In updated a restricted-distribution advisory from October. The targets: non-banking financial companies, digital lending platforms, fintechs, payment aggregators and payment service providers, wallets and cryptocurrency exchanges. Several organisations were hit again after incomplete remediation, with attacks concentrated on holidays, weekends and other low-monitoring windows.

The chain described by CERT-In:

  • initial access through internet-exposed non-production environments (UAT, staging, test);
  • move into production using credentials, tokens or network paths shared between environments;
  • code analysis and modified application packages that call payment APIs, bypassing controls that assume only the legitimate app talks to the APIs;
  • IP allow-list bypass by routing traffic through already trusted third-party services, such as content acceleration or app-store review fetchers;
  • cloud abuse (for example AWS Systems Manager and Lambda) to run commands and deploy files;
  • test transactions of about one rupee before large transfers, also through payment aggregators’ settlement APIs; at exchanges, theft of wallet keys;
  • fast exploitation of newly published vulnerabilities, with file-upload and deserialization chains leading to web shells in PHP and Node.js applications.

CERT-In reports many near-identical script variants refined in rapid succession: a pattern consistent with AI-generated code. The tools observed, sometimes combined, are ARTEX (based on PentAGI, according to Infosecurity Magazine), CyberStrikeAI and the attack search engine Cairn. According to the outlets, earlier campaigns using these tools were mostly linked to Chinese or Chinese-speaking actors; CERT-In does not attribute Crafty Falcon to a country.

In an indicative mapping to MITRE ATT&CK: T1190 (exposed applications), T1078 (valid accounts), T1505.003 (web shell), T1651 (cloud administration commands), T1090 (proxy).

Exposed ARTEX consoles

According to Censys’s 5 November report, covered by Infosecurity Magazine and TechCrunch, 1,465 hosts expose an ARTEX console; about 650 sit in two blocks of identical hosting, leaving roughly 815 distinct deployments. TechCrunch reports that they are accessible without a password and that not all are used maliciously.

On 155 hosts Censys also finds other tools classified as threats, including the Vshell, Cobalt Strike and Adaptix C2 frameworks; CyberStrikeAI appears alongside ARTEX on 70 hosts; 32 hosts have open directories with post-exploitation kits and implants. According to Censys, quoted by TechCrunch, the instance used against the Korean banks had been exposed for months and held data from attacks on payment and gaming companies as well.

The original ARTEX repository was removed in October, but the code circulates in forks and copies, including ScopeWeaver and a commercial version, ARTEX PRO. Only The Hacker News reports a backdoored version of ARTEX that exposes 62 consoles to root access.

The Korean campaign according to Hunt.io

According to Security Affairs, which covers Hunt.io’s analysis, seven South Korean financial institutions were hit, with personal, employment, income and loan data of about 66,000–67,000 people and 2,200 companies. In addition:

  • on the already known 38.244.50[.]120, the ARTEX login page answers on ports 443 and 8090;
  • a jump server in Hong Kong, 23.248.226[.]250 (AS138415), had an open directory with reverse SSH tunnel scripts, credentials for several model providers and, on port 8001, a JNDI/LDAP exploitation setup with DNS callback;
  • the operator used Claude Code connected to DeepSeek through the reseller xcai[.]pro, with 19 custom «skills», including payment logic testing and insecure direct object reference (IDOR) testing;
  • other ARTEX-related hosts: 130.94.31[.]86 (port 18888) and 156.251.11[.]39, with an SMB reconnaissance kit that writes the file artex_probe_11876.txt.

Impact for European and Italian companies

The known victims are in India and South Korea, but the entry points are common everywhere: exposed test environments, secrets shared with production, payment APIs that trust the client, unpatched web vulnerabilities. For payment institutions, e-money issuers and crypto-asset service providers subject to DORA, and for NIS2 entities, an incident of this kind triggers notification obligations.

What to do now

  1. Take UAT, staging and test off the internet or put them behind a VPN; no credentials, tokens, keys, databases or network paths shared with production.
  2. Patch exposed services fast, starting with file-upload and deserialization flaws in PHP and Node.js, with WAF virtual patching in the meantime; read-only application directories and integrity monitoring against web shells.
  3. Don’t trust the client: server-side transaction validation, requests signed with nonce and timestamp, app attestation.
  4. Hold and review micro-transfers to new beneficiaries followed by large amounts.
  5. Watch the cloud: alerts on remote command execution, serverless function changes, new roles, users and keys.
  6. Full monitoring on weekends and holidays, with SIEM correlations for machine-speed activity.
  7. Search for the indicators in firewall and proxy logs, and for traffic to language model APIs from servers that shouldn’t use them.
  8. Ban unauthorised offensive tools, ARTEX and its copies included, on corporate infrastructure.

How ARXDOME helps

ARX Expose shows what an attacker sees from outside, so a forgotten staging environment surfaces before someone finds it. ARX Sentry brings server and firewall logs into the SOC and NOC pages, at night and on weekends too, and ARX Horizon enriches events with threat lists. ARX Shield keeps track of open vulnerabilities on monitored machines.

Last updated:

Any questions?

Let’s talk about your company’s needs, with no obligation.

Contact us

Related articles