Crafty Falcon: An AI campaign that attacks payment APIs at machine speed

India’s National Cyber Security Team has issued a critical warning about a campaign called Crafty Falcon. In this campaign, attackers use publicly available AI tools to automate the entire attack chain, targeting the payment interfaces of financial firms.

What happened

On November 4, 2026, India’s CERT-In issued an advisory marked “Critical,” updating an earlier warning from October 9 regarding the exploitation of payment and payout APIs for unauthorized money transfers. Its financial team, CSIRT-Fin, named the actor “Crafty Falcon” and described it as an AI-enabled financial attacker. According to the agency, the threat is active and ongoing, and financial institutions should treat the campaign as an immediate and persistent threat.

The targets include non-bank financial companies, digital lenders, fintech firms, payment service providers, and cryptocurrency exchanges. The agency also notes that some organizations have been attacked repeatedly, including those that implemented corrective measures following the October warning.

Important note: According to available reports, the warning does not confirm that all reported incidents are necessarily the work of this particular group.

What's New in the Campaign

The brief suggests that attackers have their own “trained” models. According to the agency, this is not the case. Attackers use publicly available AI tools, including agent-based penetration testing tools, to carry out attacks autonomously and at machine speed. The core of this development is not a new weapon, but a lowering of the barrier to entry: what previously required an experienced person and a great deal of time can now be repeated over and over again by an automated tool.

In practice, this means that AI rapidly generates and modifies a large number of customized attack scripts. When one attempt fails, it tries another, slightly modified one. For defenses that rely on signatures or slow human response, this is a significantly tougher opponent. However, the term “record-breaking speed” in the headline is more of a journalistic simplification; the agency refers to attacks at machine speed.

How the attack works

According to the agency’s description, the attack involves several typical steps:

• Access via non-production environments. Attackers most often start in test, acceptance, or “staging” environments, which are accessible from the Internet and tend to be less secure than production environments.
• Cloud infrastructure. Cloud environments are also targeted, including computing and serverless services as well as management interfaces.
• Reverse engineering of applications. Attackers dissect application code to understand how they communicate with the backend and create modified installation packages capable of sending unauthorized requests. The agency cites as vulnerabilities, for example, hard-coded credentials, keys, and tokens directly within applications, or insufficient validation of addresses loaded within the application.
• Bypassing IP-based protection. Requests are sent through authorized third-party services, so lists of allowed addresses cannot detect them.
• Testing with microtransactions. Before making a larger transfer, attackers test the route with a symbolic amount, even as little as one Indian rupee.
• Exploitation of merchant settlement interfaces. Money is transferred through these interfaces directly from the victims’ own accounts.
• Timing. According to the agency, attacks occurred primarily during holidays, weekends, and periods of reduced oversight to delay detection.

What the agency recommends

The list of recommendations is long, but the core is surprisingly “boring”—and that’s exactly why it works:

• Separate test and staging environments from production, and do not share access credentials or API keys between them.
• Validate transactions on the server side, not just within the application.
• Sign API requests using one-time values and timestamps so they cannot be reused.
• Use application attestation to detect modified or repackaged installation packages, and use mutually authenticated encrypted connections with restricted keys for individual vendors.
• Strengthen monitoring of the internet access of available applications, especially on weekends and holidays.
• Restrict unauthorized AI tools on corporate systems and monitor the use of language model APIs.

What does this mean for the Slovak market?

The warning is directed at the Indian financial sector, and the specific incidents mentioned involve companies there. It does not directly affect Slovak e-shops or payment gateways. However, the lesson is universal: techniques such as unprotected staging environments, keys stored within the application, allowed IP addresses as the sole form of protection, or lax monitoring over the weekend apply in any country. The difference is that an attacker using an AI tool can test them much faster than ever before.

For payment interface developers, this is primarily a reminder that security based solely on the assumption that an attacker will be slow is no longer sufficient. Defenses must be designed to withstand a large number of automated attempts: through request signing, server-side validation, limited keys, and monitoring of unusually small transactions.

Summary

The Crafty Falcon campaign isn’t proof that “AI hackers” have a secret weapon. It demonstrates that readily available AI tools accelerate known attack techniques, especially where basic security measures have been neglected for a long time. The agency’s report is therefore particularly useful as a checklist: if you haven’t given these points much thought for a while, now is the time to do so.