Reproduce “A Behavioural Framework for Predicting and Understanding Loss of Control in Frontier Artificial Intelligence Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yao Huang, Yichi Zhang, Yifan Wang, Yitong Sun, Haoyu Dong, Jifan Ma, Jinzhe Tu, Xiaojian Li and 3 more
The authors present a framework for estimating frontier AI systems’ loss-of-control risk by combining scores for misaligned motive, harm-enabling capability, and monitoring evasion. They multiply the factor scores and test the result in contained scenarios monitored by GPT-4.1. Across 13 models, scores correlate with average simulated failure rates (r = 0.84), but this does not establish real-world loss of control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.