paper-with-me

홈 › Papers

AI Sandbagging: Language Models can Strategically Underperform on Evaluations

2024-06-11 · Teun van der Weij, Felix Hofstätter, Ollie Jaffe, Samuel F. Brown, Francis Rhys Ward

Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations to understate the AI's actual capability. These conflicting interests lead to the problem of sandbagging, which we define as strategic underperformance on an evaluation. In this paper we assess sandbagging capabilities in contemporary language models (LMs). We prompt frontier LMs, like GPT-4 and Claude 3 Opus, to selectively underperform on dangerous capability evaluations, while maintaining performance on general (harmless) capability evaluations. Moreover, we find that models can be fine-tuned, on a synthetic dataset, to hide specific capabilities unless given a password. This behaviour generalizes to high-quality, held-out benchmarks such as WMDP. In addition, we show that both frontier and smaller models can be prompted or password-locked to target specific scores on a capability evaluation. We have mediocre success in password-locking a model to mimic the answers a weaker model would give. Overall, our results suggest that capability evaluations are vulnerable to sandbagging. This vulnerability decreases the trustworthiness of evaluations, and thereby undermines important safety decisions regarding the development and deployment of advanced AI systems.

📄 PDF Abstract BibTeX arXiv:2406.07358

Code (1)

teunvdweij/sandbagging 공식 구현 pytorch

Methods 이 논문이 사용한 방법론

Attention 설명 없음
Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Layer Normalization Unlike batch normalization, Layer Normalization directly estimates the normalization statistics from the summed inputs…
Linear Layer A Linear Layer is a projection $\mathbf{XW + b}$.
BPE Byte Pair Encoding, or BPE, is a subword segmentation algorithm that encodes rare and unknown words as sequences of subword units. The intuition is that various word…
Label Smoothing Label Smoothing is a regularization technique that introduces noise for the labels. This accounts for the fact that datasets may have mistakes in them, so maximizing the…
Adam 설명 없음
Residual Connection 설명 없음

Similar Papers 제목 키워드 기반

Noise Injection Reveals Hidden Capabilities of Sandbagging Language Models

2024-12-02 · Cameron Tice, Philipp Alexander Kreer, Nathan Helm-Burger, Prithviraj Singh Shahani 외

Capability evaluations play a critical role in ensuring the safe deployment of frontier AI systems, but this role may be undermined by intentional underperformance or ``sandbagging.'' We present a novel model-agnostic me…

MMLUMultiple-choice

In-Context Environments Induce Evaluation-Awareness in Language Models

2026-03-04 · Maheep Chaudhary arxiv

Humans often become more self-aware under threat, yet can lose self-awareness when absorbed in a task; we hypothesize that language models exhibit environment-dependent \textit{evaluation awareness}. This raises concerns…

Code Generation

LLMs Can Covertly Sandbag on Capability Evaluations Against Chain-of-Thought Monitoring

2025-07-31 · Chloe Li, Mary Phuong, Noah Y. Siegel arxiv

Trustworthy evaluations of dangerous capabilities are increasingly crucial for determining whether an AI system is safe to deploy. One empirically demonstrated threat is sandbagging - the strategic underperformance on ev…

Below-Chance Blindness: Prompted Underperformance in Small LLMs Produces Positional Bias Rather than Answer Avoidance

2026-04-28 · Jon-Paul Cacioli arxiv

Detecting sandbagging--the deliberate underperformance on capability evaluations--is an open problem in AI safety. We tested whether symptom validity testing (SVT) logic from clinical malingering detection could identify…

Auditing Games for Sandbagging

2025-12-08 · Jordan Taylor, Sid Black, Dillon Bowen, Thomas Read 외 arxiv

Future AI systems could conceal their capabilities ('sandbagging') during evaluations, potentially misleading developers and auditors. We stress-tested sandbagging detection techniques using an auditing game. First, a re…