Retail bank onboarding chatbot served one user another user's KYC document
A US retail bank's onboarding chatbot returned a partial KYC document from another applicant during a brief retrieval-layer misconfiguration. The exposure window was 4 hours.
Retrieval boundaries are identity boundaries. When they slip, the failure is a data event, not an AI event.
Key facts
- What
- A US retail bank's onboarding chatbot returned a partial KYC document from another applicant during a brief retrieval-layer misconfiguration.
- Incident date
- Feb 14, 2026
- Who
- Anonymized: Retail Bank · US · $300B+ assets
- Failure mode
- Data Leakage
- AI surface
- Chatbot
- Severity
- High
What happened
A US retail bank's onboarding chatbot returned content from another applicant's KYC submission for a four-hour window in early 2026 after a retrieval-layer change pushed during a Friday deployment. The bank disclosed to the affected user and the state attorney general. No regulatory action followed, but the bank rebuilt its retrieval boundary and added an identity-bound retrieval check.
The case is anonymized. The pattern is what every retail bank with an AI customer surface should expect to see at least once.
What broke inside the model
- 01 · TriggerA request triggers retrieval or context loading.
- 02 · Model stepThe context pulls in another user's content.
- 03 · Control gapNo boundary enforces isolation at the moment of output.
- 04 · FailurePrivate data crosses into the response.
- 05 · ConsequenceOne user sees another's data, and disclosure follows.
One user's content crosses the retrieval boundary into another's response.
Retrieval boundary failure. The retrieval layer surfaced content from another tenant's context because the deployment's identity-binding broke during a release. The model did what the retrieval layer told it to do. The retrieval layer was wrong.
What it cost
Sources
- Customer-DisclosedRealm Labs case file under NDAfailureindex.ai
Cite this entry
https://failureindex.ai/failures/anonymized-retail-banking-onboarding-leakageAI Failure Index. "Retail bank onboarding chatbot served one user another user's KYC document" (FI-0022). Realm Labs. https://failureindex.ai/failures/anonymized-retail-banking-onboarding-leakage (indexed May 13, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0022. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
OmniGuard's identity-bound enforcement reads each piece of retrieved content against the current user's identity at the moment the model is about to use it. Content that does not belong to the current user is dropped before the model sees it. The four-hour window becomes a zero-hour window.