Skip to content
AI.info

The Pulse

Researcher says CreditNirvana flaws exposed 6.6 million accounts

Two authentication flaws exposed borrower accounts and about 315,000 recorded collection calls, according to researcher Eaton Zveare. CreditNirvana fixed the vulnerabilities in June, before the detailed findings were published in November.

Researcher says CreditNirvana flaws exposed 6.6 million accounts

Two authentication flaws in CreditNirvana’s AI debt collection systems exposed 6.6 million borrower accounts and about 315,000 recorded calls, according to security researcher Eaton Zveare. His November 18 disclosure describes how an unauthenticated visitor could reach both an AI call center and the main collections platform.

CreditNirvana, an India-based debt collection platform owned by Perfios, fixed the vulnerabilities in June. The findings describe information that was accessible while the flaws existed, not proof that someone else accessed or stole it.

Calls and AI agents

The call-center flaw involved an application programming interface key embedded in JavaScript sent to a user’s browser. Zveare found that the system did not use the token returned by Amazon Cognito, the service handling its login, and that he could bypass the sign-in process.

The approximately 315,000 recordings belonged to Aditya Birla Capital Finance customers, the researcher said. The system also held transcriptions and analysis of the calls, along with recommendations for actions after them.

The accessible functions extended beyond reviewing calls. Zveare said someone with access could build and manage AI-powered campaigns and browse or manage the AI agents available in the collection system.

Accounts and documents

The main platform had a separate authentication failure that allowed an administrator login, according to the researcher. He found that the platform did not check the Keycloak token sent with a login request and instead relied on a username in the request.

An initial test also returned a list of more than 10,000 users without authentication, Zveare said. The Aditya Birla Capital Finance instance contained more than 3.4 million accounts; the other affected instances, for BharatPe and International Asset Reconstruction Company, held an additional 3.2 million.

The exposed accounts contained sensitive information including identity cards, credit reports, legal documents and records of communications, according to the disclosure. The platform also provided access to call recordings and AI summaries, legal case management and the location tracking of debt collection agents.

The researcher found a feature for requesting legal action on an account, as well as administrative access to SlashRTC, a contact-center and cloud communication provider. He said it was unclear whether the platform could be used to cancel a debt.

Reporting and fixes

Zveare reported the vulnerabilities through India’s Computer Emergency Response Team, CERT-In, on May 30, 2026, and supplied a proof-of-concept video on June 1. Both vulnerabilities were fixed on June 2, the same day CreditNirvana and Perfios became aware of them, he said.

A vulnerability note and identifiers CVE-2026-110474 and CVE-2026-110475 followed on October 26. CERT-In rated the call-center issue high severity at 8.7 and the main-platform issue critical at 9.3.

It remains unclear how long the exposure lasted or whether anyone exploited the flaws before the fix, DiffVibe reported. WIRED reported that CreditNirvana and Perfios did not return its requests for comment.

Sources

  1. Original source Eaton Works eaton-works.com

Other pages read

  1. WIRED wired.com
  2. DiffVibe diffvibe.com

Related stories

Explore

More articles