Category: AI Agent Security | Reading time: 3 min
Security researcher Eaton Zveare found two bugs in CreditNirvana, an AI debt-collection platform owned by Indian fintech Perfios, that exposed about 6.6 million borrower accounts and roughly 315,000 call recordings. WIRED published the findings on November 18. The flaws were patched in June and disclosed in October.
What was exposed
CreditNirvana describes its agents as able to "decide the next best action, and execute it autonomously." According to WIRED, the exposed data included credit reports, ID cards, legal documents, and emails, texts and call transcripts. The platform's three corporate customers are prominent Indian financial services firms.
Control of the agents
The exposure went beyond data. One high-severity bug could have let an attacker manage the platform's AI agents and run their own collection campaigns. A critical bug also allowed tracking the locations of human collectors.
Why it matters for regulated enterprises
When an autonomous agent contacts borrowers on a lender's behalf, the lender needs to know that every action came from an authorized agent following its instructions. A flaw that lets an outsider take over agent campaigns means an institution may not be able to tell its own agent's actions from an attacker's.
Firms using third-party agent platforms in collections, servicing or other regulated workflows should ask vendors how agent identity is enforced, who can change agent behavior, and what verifiable record exists of each action taken.
Source: WIRED