FairStress tests whether AI changes its decisions under pressure
The Allen Institute for AI introduced FairStress on October 20. Its 300 decision scenarios cover 29 domains. The benchmark asks whether matched arguments change a model's choice differently when candidates' demographic identities change.
Verified 12:04 AM PDT · 2 original sources
These are the authors' research results, not an independent audit of real hiring or lending. The benchmark tests one definition of fairness, based on equal criteria and equal resistance to persuasion.
Watch independent reruns on current models and real workflows. An unchanged first answer can hide a decision that becomes unstable when challenged.
Audit the story
Original sources
Company claims remain company claims. Follow the reporting and judge the evidence directly.
- Allen Institute for AI, via Hugging FaceFairStress: A deeper evaluation of fairness in LLM decision-making ↗
- Allen Institute for AI and coauthorsAsymmetrically Unfair Decision Making and Reasoning in LLMs ↗
Continue the edition