How AI-driven cyberattacks are targeting India's financial sector

The Indian Computer Emergency Response Team (CERT-In) has warned of escalating AI-driven cyberattacks on financial institutions, with hackers increasingly exploiting their own payment infrastructure to initiate unauthorized transactions.

In an advisory issued on 4 November, CERT-In identified the threat actor as “Crafty Falcon” and classified the campaign as critical. The warning coincides with the agency's new AI Security Auditing Guidelines, which establish a risk-based framework for assessing the security of artificial intelligence systems.

Mint explains.

What exactly is Craft Falcon?

Craft Falcon is an AI-assisted financial threat actor identified by CERT-In's Computer Security Incident Response Team-Finance (CSIRT-Fin). Its targets include banks, non-banking financial companies (NBFCs), digital lenders, fintech firms, payment aggregators and service providers, and cryptocurrency exchanges.

Attackers use AI to identify vulnerabilities, execute attacks at machine speed and exploit institutions' payment and disbursement application programming interfaces (APIs), according to the agency.

The latest advisory updates a 9 October warning. CERT-In said attacks continued even against entities that had received restricted advisories and undertaken remediation.

How are attackers using AI to breach financial systems?

CERT-In has identified several publicly available offensive AI tools used in the campaign, including the ARTEX family of large language model-based tools, PentAGI, an autonomous multi-agent AI penetration-testing system, the Cairn attack search engine and CyberStrikeAI.

These tools can autonomously execute attack chains, generate and refine scripts, and adapt payloads to responses from targeted systems. Attackers also rotate internet protocol (IP) addresses within hours or days to evade detection and blocking.

What vulnerabilities are attackers exploiting in payment systems?

Attackers target APIs used for payments, loan disbursements and merchant settlements. Compromising these interfaces allows them to attempt unauthorized transfers through financial institutions' own infrastructure.

Other targets include cloud infrastructure, exposed testing environments, shared credentials and access tokens. Techniques include reverse-engineering application code, modifying application packages and bypassing IP restrictions through whitelisted third-party services.

Attackers also use micro-value transactions, sometimes as small as ₹1, to test whether unauthorized fund transfers work before attempting larger transactions.

Which other financial services are being targeted?

Cryptocurrency exchanges and virtual digital asset service providers have reported incidents involving stolen wallet keys and siphoned funds. Compromised merchant settlement APIs at payment aggregators can similarly facilitate unauthorized transfers to beneficiaries.

CERT-In also warned that attackers have retargeted entities after incomplete remediation, sometimes timing attacks to coincide with holidays and weekends, when monitoring may be limited.

What are CERT-In's new AI Security Auditing Guidelines?

The 107-page guidelines establish a framework for assessing AI security and governance throughout the system lifecycle, from development and procurement to deployment and decommissioning.

They cover nine audit domains: AI governance, data security and privacy, model risk management, fairness, explainability, cybersecurity, continuous monitoring, human oversight and third-party accountability.

Specific requirements for generative AI, large language models, retrieval-augmented generation and agentic AI address risks such as prompt injection, data poisoning, sensitive-information disclosure and unauthorized tool use.

How will the guidelines classify AI systems?

The framework defines four risk categories: low, moderate, high and critical, based on eight factors including data sensitivity, potential impact, system autonomy, external exposure and complexity.

The category determines audit requirements. Low-risk systems can undergo self-assessment or basic internal reviews. Moderate-risk systems require documented internal audits and independent audits by CERT-In-empanelled organisations at least annually. High-risk systems require periodic independent third-party audits, while critical systems require independent audits and continuous monitoring.

What will financial institutions need to do?

The guidelines call for AI-specific vulnerability assessments, adversarial testing, API security checks, model-integrity verification and continuous monitoring. They also prescribe forensic readiness and 180-day log retention. Incidents meeting applicable CERT-In reporting criteria must be reported within six hours.

Financial institutions will need to strengthen cybersecurity teams, regularly test AI systems and patch vulnerabilities before attackers exploit them.

Can financial institutions rely on CERT-In alone to tackle these threats?

Industry experts say government-led cybersecurity initiatives must be backed by stronger internal capabilities at banks and fintech companies.

“The frequency of AI-driven attacks on banking and financial infrastructure is only going to accelerate. The flurry of attacks last month tells a telling tale. Elon Musk captured it well when he said that, with AI, every day is a zero day,” said Sharat Chandra, founder of EmpowerEdge Ventures, a venture-building firm for fintechs and startups.

Chandra said CERT-In’s AI security auditing guidelines were a welcome step, while its access to Anthropic’s Mythos could improve its ability to detect attacks and vulnerabilities early. However, he said banks, NBFCs and fintech companies needed to strengthen their own cybersecurity capabilities, as shrinking response windows left them with less time to identify and patch vulnerabilities.

What risks do AI-driven cyberattacks pose to customers, and how can they protect their money?

Customers risk account takeovers, unauthorized transactions and financial losses if attackers access their credentials or digital sessions.

Experts said cybercriminals were increasingly using AI to automate and personalise credential theft and session hijacking, potentially bypassing traditional security controls.

The recent Stockal incident, in which customer accounts were compromised and trades executed without consent, illustrates the risks of unauthorized access to financial accounts, they said.

Customers should use strong, unique passwords, enable two-factor authentication (2FA) and never share passwords, one-time passwords (OTPs) or other authentication details. Regularly checking accounts for suspicious activity is also essential, Chandra said.