Deceptive Alignment Monitoring

Andres Carranza, Dhruv Bhandarkar Pai, Rylan Schaeffer, Arnuv Tandon, Sanmi Koyejo

Keywords: Interpretability, Deceptive Alignment, Deep Learning, Scalable Oversight, ELK

TL;DR: We propose a new paradigm of adversarial machine learning, deceptive alignment monitoring, in which mechanistically anomalous model behavior serves as a basis fo model misalignment, and propose aa variety of new research directions in the field.

Abstract: As the capabilities of large machine learning models continue to grow, and as the autonomy afforded to such models continues to expand, the spectre of a new adversary looms: the models themselves. The threat that a model might behave in a seemingly reasonable manner, while secretly and subtly modifying its behavior for ulterior reasons is often referred to as deceptive alignment in the AI Safety & Alignment communities. Consequently, we call this new direction Deceptive Alignment Monitoring. In this work, we identify emerging directions in diverse machine learning subfields that we believe will become increasingly important and intertwined in the near future for deceptive alignment monitoring, and we argue that advances in these fields present both long-term challenges and new research opportunities. We conclude by advocating for greater involvement by the adversarial machine learning community in these emerging directions.

Submission Number: 84