Adversarial Multiplicity: Cherry-Picked Models and the Limits of Disclosure
Factual evidence
What the source reports
Researchers introduced adversarial multiplicity, demonstrating how institutions or AI agents can cherry-pick equally accurate models to manipulate outcomes.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.LG — Machine Learning · RESEARCH
- Published by source
- 6 November 2026
- Collected by OneBench
- 7 Nov 2026, 02:02 UK
Stored source excerpt
arXiv:2611.07378v1 Announce Type: new Abstract: Predictive multiplicity -- where similarly accurate models disagree on individual predictions -- can be leveraged to optimize secondary metrics (e.g.,…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Equally accurate models can yield conflicting individual credit or underwriting decisions, exposing automated pipelines to deliberate bias and gaming.
Do what
Review model risk management frameworks to test candidate models for prediction multiplicity before approving automated decisioning pipelines.