# Exploring the Robustness of Model-Graded Evaluations and Automated Interpretability

arXiv: https://arxiv.org/abs/2312.03721 · HTML: https://mlsanity.com/paper/2312.03721/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 9 of its 16 references that are in the map.
- First posted on arXiv 26 Nov 2023.

## Builds on (most cited first, 9 of 9)

- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting](https://mlsanity.com/paper/2305.04388/) (2023)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://mlsanity.com/paper/2212.09251/) (2022)
- [Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark](https://mlsanity.com/paper/2304.03279/) (2023)
- [Exploring the MIT Mathematics and EECS Curriculum Using Large Language Models](https://mlsanity.com/paper/2306.08997/) (2023)
- [Evaluating Shutdown Avoidance of Language Models in Textual Scenarios](https://mlsanity.com/paper/2307.00787/) (2023)

## Built on by (most cited first, 2 of 2)

- [Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight Systems](https://mlsanity.com/paper/2504.07831/) (2025)
- [Harmonic LLMs are Trustworthy](https://mlsanity.com/paper/2404.19708/) (2024)

## Abstract (arXiv)

There has been increasing interest in evaluations of language models for a variety of risks and characteristics. Evaluations relying on natural language understanding for grading can often be performed at scale by using other language models. We test the robustness of these model-graded evaluations to injections on different datasets including a new Deception Eval. These injections resemble direct communication between the testee and the evaluator to change their grading. We extrapolate that future, more intelligent models might manipulate or cooperate with their evaluation model. We find significant susceptibility to these injections in state-of-the-art commercial models on all examined evaluations. Furthermore, similar injections can be used on automated interpretability frameworks to produce misleading model-written explanations. The results inspire future work and should caution against unqualified trust in evaluations and automated interpretability.