Security
AI Debt Collection Platform Exposed 6.6M Accounts in India
Two security flaws in CreditNirvana left borrower data, credit reports, and call recordings accessible for years before June 2024 patches.
Key takeaways
- CreditNirvana, an AI debt collection platform used by major Indian fintech firms, exposed 6.6 million borrower accounts through two security vulnerabilities patched in June 2024.
- The critical flaw allowed potential attackers to impersonate administrative users and access credit reports, ID documents, court records, and millions of communications.
- A second vulnerability exposed 315,000 call recordings and could have enabled unauthorized control of AI collection agents.
- India's CERT-In classified the flaws as high and critical severity; the vulnerabilities may have existed for years before discovery.
- The incident highlights security risks as AI systems increasingly aggregate and act autonomously on sensitive financial data.
Security flaws in AI collections system
Two vulnerabilities in CreditNirvana, an AI-driven debt collection platform operated by Indian fintech company Perfios, exposed 6.6 million borrower accounts and approximately 315,000 call recordings, according to security research disclosed this week. The platform markets itself as an autonomous system that analyzes borrower context and executes collection actions without human intervention.
Independent researcher Eaton Zveare discovered the flaws and reported them through India's Computer Emergency Response Team (CERT-In). The vulnerabilities were patched in June, with public disclosure following in October. CERT-In classified one flaw as high severity and the other as critical.
The exposed data included credit scores, identification documents, court records, case management details, and communications such as emails, texts, and call transcripts related to unpaid accounts. The vulnerabilities may have existed in CreditNirvana's systems for years before discovery.
Two distinct attack vectors
The high-severity vulnerability affected the platform's call center infrastructure, exposing hundreds of thousands of recorded calls and AI-generated recommendations for follow-up actions. An attacker exploiting this flaw could have manipulated AI agents within the system to launch unauthorized collection campaigns.
The critical vulnerability impacted CreditNirvana's core platform and specifically compromised data from three corporate customers—all major Indian financial services firms. According to CERT-In's assessment, successful exploitation would have allowed attackers to enumerate users, establish sessions as any user including administrators, and access or modify sensitive information with elevated privileges. The flaw also exposed the physical location tracking of human debt collection agents.
Why it matters
As financial institutions increasingly deploy AI systems that aggregate and act on sensitive personal data, this incident illustrates the amplified risk when security failures occur. Debt collection data is particularly sensitive because it documents financial distress, unpaid obligations, and often includes extensive personal histories that borrowers would not willingly disclose. The autonomous nature of CreditNirvana—making collection decisions and taking action without human oversight—means a single platform vulnerability can expose the financial records of millions across multiple institutions. For enterprises evaluating AI vendors, this case underscores the need for rigorous third-party security assessments, especially when systems handle regulated financial data.
Rapid response, lingering questions
Zveare noted that Perfios resolved the vulnerabilities within one day of notification, demonstrating responsive remediation. However, the researcher emphasized the scale of potential exposure given CreditNirvana's use by some of India's largest fintech companies for autonomous debt collection operations.
Neither CreditNirvana nor Perfios responded to requests for comment on the incident, the duration the vulnerabilities existed, or whether any unauthorized access occurred before patching.
The details were first reported by WIRED.