# Adaptive auditing of AI systems with anytime-valid guarantees

arXiv: https://arxiv.org/abs/2605.07002 · HTML: https://mlsanity.com/paper/2605.07002/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 29 of its 51 references that are in the map.
- First posted on arXiv 7 May 2026.

## Builds on (most cited first, 29 of 29)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned](https://mlsanity.com/paper/2209.07858/) (2022)
- [SPECTER: Document-level Representation Learning using Citation-informed Transformers](https://mlsanity.com/paper/2004.07180/) (2020)
- [Domino: Discovering Systematic Errors with Cross-Modal Embeddings](https://mlsanity.com/paper/2203.14960/) (2022)
- [Safe Testing](https://mlsanity.com/paper/1906.07801/) (2019)
- [FairVis: Visual Analytics for Discovering Intersectional Bias in Machine Learning](https://mlsanity.com/paper/1904.05419/) (2019)
- [Anchor Points: Benchmarking Models with Much Fewer Examples](https://mlsanity.com/paper/2309.08638/) (2023)
- [Consistency of random forests](https://mlsanity.com/paper/1405.2881/) (2014)
- [Red-Teaming for Generative AI: Silver Bullet or Security Theater?](https://mlsanity.com/paper/2401.15897/) (2024)
- [Hypothesis testing with e-values](https://mlsanity.com/paper/2410.23614/) (2024)
- [Automated Data Slicing for Model Validation:A Big data - AI Integration Approach](https://mlsanity.com/paper/1807.06068/) (2018)
- [There are natural scores: Full comment on Shafer, "Testing by betting: A strategy for statistical and scientific communication"](https://mlsanity.com/paper/2102.05569/) (2021)
- [Universal Inference](https://mlsanity.com/paper/1912.11436/) (2019)
- [AutoBencher: Towards Declarative Benchmark Construction](https://mlsanity.com/paper/2407.08351/) (2024)
- [Adaptive Testing of Computer Vision Models](https://mlsanity.com/paper/2212.02774/) (2022)
- [E-detectors: a nonparametric framework for sequential change detection](https://mlsanity.com/paper/2203.03532/) (2022)
- [Query-Efficient Black-Box Red Teaming via Bayesian Optimization](https://mlsanity.com/paper/2305.17444/) (2023)
- [Active Fairness Auditing](https://mlsanity.com/paper/2206.08450/) (2022)
- [E-valuator: Reliable Agent Verifiers with Sequential Hypothesis Testing](https://mlsanity.com/paper/2512.03109/) (2025)
- [Scaling Up Active Testing to Large Language Models](https://mlsanity.com/paper/2508.09093/) (2025)
- [General Agent Evaluation](https://mlsanity.com/paper/2602.22953/) (2026)
- [Is this model reliable for everyone? Testing for strong calibration](https://mlsanity.com/paper/2307.15247/) (2023)
- [Audit Me If You Can: Query-Efficient Active Fairness Auditing of Black-Box LLMs](https://mlsanity.com/paper/2601.03087/) (2026)
- [A Brief Tutorial on Sample Size Calculations for Fairness Audits](https://mlsanity.com/paper/2312.04745/) (2023)
- [When the Domain Expert Has No Time and the LLM Developer Has No Clinical Expertise: Real-World Lessons from LLM Co-Design in a Safety-Net Hospital](https://mlsanity.com/paper/2508.08504/) (2025)
- [Active Hypothesis Testing under Computational Budgets with Applications to GWAS and LLM](https://mlsanity.com/paper/2512.01423/) (2025)

## Built on by (most cited first, 3 of 3)

- [Pay Only for Disagreement: Certified No-Regression Verdicts for Model Updates with Matching Label-Complexity Bounds](https://mlsanity.com/paper/2609.17560/) (2026)
- [Black-Box Auditing of Epistemic Reliability in Multi-Agent Debate Distillation](https://mlsanity.com/paper/2609.32361/) (2026)
- [Learning to Detect Shifts in Unstructured Data Streams via Monitoring-by-Betting](https://mlsanity.com/paper/2610.17399/) (2026)

## Abstract (arXiv)

A major bottleneck in characterizing the failure modes of generative AI systems is the cost and time of annotation and evaluation. Consequently, adaptive testing paradigms have gained popularity, where one opportunistically decides which cases and how many to annotate based on past results. While this framework is highly practical, its extreme flexibility makes it difficult to draw statistically rigorous conclusions, as it violates classical assumptions: the number of observations is typically limited (often 10 to 50 cases) and decisions regarding sampling and stopping are made in the midst of data collection rather than based a pre-specified rule. To characterize what statistical inferences can be drawn from highly adaptive audits, we introduce a hypothesis testing framework from two 'dueling' perspectives: (i) the model's null that asserts there is no failure mode with performance below a target threshold versus (ii) the auditor's null that asserts they have a sampling strategy that will uncover a failure mode. Leveraging Safe Anytime-Valid Inference (SAVI), we formalize the auditor as conducting 'testing by betting', which translates into simultaneous e-processes for testing the dueling null hypotheses. Furthermore, if the auditor is sufficiently powerful, we prove that these two hypotheses are asymptotically inverses of each other, in that passage of a stringent audit does in fact certify the AI system as being globally robust. Empirically, we demonstrate that our proposed testing procedures maintain anytime-valid type-I error control, outperform pre-specified testing methods, and can reach statistically rigorous conclusions sometimes with as few as 20 observations.