CERT-In warns crypto exchanges of escalating AI-assisted attacks

India’s national cybersecurity agency, the Computer Emergency Response Team (CERT-In), designated the AI-assisted threat actor “CRAFTY FALCON” on Wednesday, warning that an active campaign against cryptocurrency exchanges and other financial firms includes successful attacks, stolen wallet keys and drained funds. CERT-In said CSIRT-Fin continues to receive reports of successful attacks, although the agency named no affected exchange and quantified no losses.

The target set

CERT-In first warned about the campaign in October, then expanded its latest advisory to identify crypto exchanges among the targets and to detail the AI tools in use. The actor has targeted nonbank lenders, digital lending platforms, fintech companies, payment aggregators and service providers, as well as crypto exchanges.

• Vulnerable applications and exposed testing environments provided entry points; shared credentials and access tokens then opened paths into production systems.
• Attackers targeted cloud infrastructure to deploy malicious files, reverse-engineered application code to build tailored exploits and used AI to generate and refine scripts quickly.
• Some operations ran on weekends and holidays to delay detection.
• Small test transactions checked whether unauthorized transfers worked before attackers moved larger sums.
• The advisory also records stolen cryptocurrency wallet keys and funds drained from wallets.

The persistence matters because CERT-In said successful attacks continued among nonbank financial companies and fintech entities that had received earlier restricted advisories and completed remediation. The operational implication is direct: remediation did not remove the risk of a recurring campaign.

AI-assisted execution

CERT-In said the AI-assisted financial threat actor “surveys its targets with machine intelligence, operates at machine speed and strikes by invoking the victim’s own payment and disbursement APIs.” The practical distinction is speed and adaptation: AI helps produce scripts, refine exploits and tailor attacks to code, while the actor still relies on stolen credentials, exposed systems and abused APIs.

In separate October research, cybersecurity firm CrowdStrike reported that an unidentified attacker combined ARTEX, an AI penetration-testing tool that also appears in CERT-In’s advisory, with Claude Code and other AI models to breach South Korean financial firms and steal data. CrowdStrike did not name that attacker in the report.

The coverage also linked a separate report saying Binance traced a $6.3 billion hack to fake job interviews and a compromised AI tool.

What does this change for crypto exchanges?

For exchange operators, the immediate risk spans wallet-key custody, production access and transaction APIs. The reported chain moves from exposed test environments and shared credentials into production, then uses payment and disbursement APIs to stage transfers. Small test transactions matter because attackers use them to validate an unauthorized-transfer path before moving larger sums.

That changes day-to-day operations. CERT-In urged at-risk companies to treat this campaign as “active and escalating,” maintain monitoring around the clock, increase staffing and lower alert thresholds during holidays, weekends and other low-coverage windows.

Michael Coates, chief information security officer at the Solana Foundation, previously told Cointelegraph that legitimate security teams need access to the strongest AI models because attackers will obtain tools capable of assisting offensive work. Ari Redbord, TRM Labs’ head of policy, said in July that attack techniques now move faster than a launch-day audit can account for, supporting continuous security reviews.

Cointelegraph also sought comment from CERT-In, CoinDCX, CoinSwitch, WazirX, ZebPay and Mudrex. The advisory did not identify any of those exchanges as an affected business.

The next operational checkpoint is the coming weekend or holiday: exchanges and financial firms need round-the-clock monitoring, added staffing and lower alert thresholds in place before those low-coverage windows, because CERT-In has classified the campaign as active, ongoing and escalating.