India warns of AI-driven cyber-attacks targeting financial firms

The rise of Crafty Falcon

The Indian Computer Emergency Response Team (CERT-In) has issued a warning regarding a new cyber threat actor known as Crafty Falcon. This group is specifically targeting non-banking financial companies, including digital lending platforms, payment aggregators, and cryptocurrency exchanges. The agency noted that these organizations have reported repeated attacks despite receiving previous warnings about similar tactics.

Crafty Falcon distinguishes itself by using AI and agentic AI tools to conduct penetration testing and breach systems. Agentic AI refers to systems capable of performing multi-step tasks autonomously to achieve a goal. The actor uses these tools to execute complete attack chains at machine speed, allowing them to adapt their payloads based on the specific responses of the target's security systems.

While these AI-powered offensive tools are publicly available, the patterns observed by CERT-In suggest a high level of sophistication. The agency observed the use of several specific tools, including CyberStrikeAI and the attack search engine Cairn. Many of the previous campaigns involving these specific tools have been linked to China-based hacking groups, though Crafty Falcon is currently being treated as a previously undocumented actor.

Exploiting the development pipeline

The attack lifecycle typically begins in non-production environments, such as user acceptance testing (UAT) or staging areas. Because these environments are often less strictly monitored than live production systems, they provide an easier entry point for hackers. Once inside, Crafty Falcon moves into production systems by exploiting shared credentials, authentication tokens, or direct network paths between the two environments.

To create effective exploits, the actor reverse-engineers the organization's application codebases. They then use modified application packages to initiate unauthorized calls to application programmable interfaces (APIs). APIs are the sets of rules that allow different software applications to communicate with each other, and they are commonly used for payment and disbursement transactions.

The actor also employs clever methods to bypass IP allow-listing, which is a security measure that only permits traffic from trusted addresses. They send their data through approved third-party fetcher pools, including those used for content acceleration. Because these common services allow any person on the internet to request specific URLs, the attacker can make their traffic appear to come from addresses the organization already trusts.

Cloud and crypto targets

Crafty Falcon specifically targets a victim's cloud infrastructure, including compute, serverless, and management services. They abuse cloud management services like AWS Systems Manager and Lambda to execute commands and deploy malicious files. By doing this, the malicious activity is able to blend in with legitimate cloud management operations, making it harder for security teams to spot.

When the goal is to move funds, the actor performs tiny test transfers, such as amounts of Rs. 1. These small transactions are used to confirm that unauthorized movement of money is possible before the actor attempts a large-value transfer. They often take advantage of merchant settlement APIs used by payment aggregators to move the stolen money.

Cryptocurrency exchanges and virtual digital asset service providers are also high-priority targets for the group. The actor focuses on stealing wallet keys to siphon funds directly from cryptocurrency wallets. Additionally, they exploit newly published vulnerabilities as soon as they become public, including file-upload and deserialization exploit chains to deploy web shells in PHP and Node.js applications.

Mitigation and monitoring

CERT-In has provided a comprehensive list of recommendations to help organizations defend against these AI-driven threats. A primary recommendation is the strict separation of UAT, staging, and testing environments from production systems. Organizations must ensure there are no shared credentials, tokens, API keys, or direct network paths between these environments to prevent lateral movement.

Security teams are urged to maintain heightened 24/7 monitoring, especially during holidays and weekends when monitoring is typically reduced. They should centralize live tracking of cloud audit logs and set alerts for high-risk events like remote command execution or the creation of new access keys. The agency also suggests treating all client applications and packages as untrusted, enforcing server-side transaction validation and API request signing.

To counter the specific use of AI by attackers, CERT-In advises monitoring and governing LLM API usage. Traffic originating from servers or non-development hosts may indicate an attack infrastructure or a planted operator console. Organizations should also adjust their security information and event management (SIEM) systems to identify machine-speed attacks and parallel attack threads that occur outside of normal business hours.