Upstart AI lending model shows racial approval disparities during independent monitorship
An independent monitor discovered racial approval disparities in Upstart's AI-driven lending model. The issues were linked to the use of educational data acting as a proxy for race, prompting a multi-year monitorship and remediation efforts.
Records by entity: Upstart Network
Upstart's AI lending model was found to produce racial approval disparities due to the use of educational data as a proxy for race.
Educational criteria in the AI model operated as a proxy for race, leading to disparate approval rates for Black applicants.
Key facts
- What
- An independent monitor discovered racial approval disparities in Upstart's AI-driven lending model.
- Incident date
- Dec 1, 2020
- Who
- Upstart Network, Inc.
- Failure mode
- Hallucination
- AI surface
- Algorithmic Decision
- Severity
- Medium
What happened
In December 2020, Upstart entered into a fair lending monitorship agreement with the NAACP LDF and SBPC. An independent monitor found that the AI model produced approval disparities for Black applicants. Upstart implemented most of the monitor's recommendations but disagreed on a specific less discriminatory alternative model.
What broke inside the model
The AI model utilized non-traditional data, including educational criteria, which functioned as a proxy for protected classes. This mechanism resulted in algorithmic bias that negatively impacted approval rates for Black applicants.
What it cost
Sources
- PressLDF, SBPC, and Upstart Announce Final Monitorship Report on AI and Fair Lendingnaacpldf.org
- Court FilingFair Lending Monitorship of Upstart Network’s Lending Modelrelmanlaw.com
Cite this entry
https://failureindex.ai/failures/upstart-lending-shows-racial-approval-disparitiesAI Failure Index. "Upstart AI lending model shows racial approval disparities during independent monitorship" (FI-0493). Realm Labs. https://failureindex.ai/failures/upstart-lending-shows-racial-approval-disparities (indexed Jun 10, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0493. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.