Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2602.05532 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908815450963968 |
|---|---|
| author | Dietz, Florian Wale, William Gilg, Oscar McCarthy, Robert Michalak, Felix Danon, Gustavo Ewbank Rodrigues de Guzman, Miguelito Klakow, Dietrich |
| author_facet | Dietz, Florian Wale, William Gilg, Oscar McCarthy, Robert Michalak, Felix Danon, Gustavo Ewbank Rodrigues de Guzman, Miguelito Klakow, Dietrich |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_05532 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities Dietz, Florian Wale, William Gilg, Oscar McCarthy, Robert Michalak, Felix Danon, Gustavo Ewbank Rodrigues de Guzman, Miguelito Klakow, Dietrich Artificial Intelligence Machine Learning 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. |
| title | Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities |
| topic | Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2602.05532 |