Evidence · DiggingBeagle record
Reliable AI-Assisted Decisions via Selective Auditing
Original statistical selective-auditing objective and experimental safety limitations.
- Published
- Oct 22, 2026
- Source role
- primary disclosure
Evidence record
Original statistical selective-auditing objective and experimental safety limitations.
Claim-level citations (3)
- supportsSelective auditing can bound aggregate AI errors escaping review, not a specific harmful approval: The paper proposes online randomized selective auditing with confidence-sensitive sampling and importance weighting to control an aggregate fraction of AI mistakes escaping review.
Abstract; problem formulation; algorithm
- supportsSelective auditing can bound aggregate AI errors escaping review, not a specific harmful approval: Reported guarantees are asymptotic or distributional and experiments include finite-sample misses; they do not prove any individual irreversible decision is correct.
Guarantees, experiments and limitations
- supportsSelective auditing can bound aggregate AI errors escaping review, not a specific harmful approval: The authors evaluate selective auditing on harmful-request screening, qualitative text annotation and data labeling; they do not report a deployed financial or publishing approval system.
Abstract, final sentences