Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities
Abstract
Detecting misalignment in large language models is challenging because models may learn to conceal misbehavior during training. Standard auditing techniques fall short: black-box methods often cannot distinguish misaligned outputs from benign ones, and mechanistic interpretability does not scale with model capabilities. We introduce Split Personality Training (SPT), which fine-tunes a second ``honest persona'' into LoRA parameters that remain inactive during normal operation. After the main model responds, we activate the LoRA adapter and insert a trigger string, enabling the honest persona to review the response while accessing the main model's latent states. We test our method on the Anthropic Auditing Game Model Organism, a benchmark where Llama-3.3-70B is trained to exploit reward hacks while concealing this behavior. SPT achieves 96% overall accuracy, whereas Anthropic reports near 0% accuracy. The honest persona reveals latent knowledge inaccessible to external observers, such as the fictional biases the compromised model was trained on.
Keywords
Cite
@article{arxiv.2602.05532,
title = {Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities},
author = {Florian Dietz and William Wale and Oscar Gilg and Robert McCarthy and Felix Michalak and Gustavo Ewbank Rodrigues Danon and Miguelito de Guzman and Dietrich Klakow},
journal= {arXiv preprint arXiv:2602.05532},
year = {2026}
}