Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.28639 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911725104660480 |
|---|---|
| author | Ramnauth, Rebecca Scassellati, Brian |
| author_facet | Ramnauth, Rebecca Scassellati, Brian |
| contents | Instruction-based suppression is widely used to prevent language models from generating prohibited content, yet it remains unclear whether suppression reduces internal representation or merely suppresses expression. We investigate this question through representational probing, attention analysis, and behavioral semantic leakage experiments across multiple transformer models. We find that prohibited concepts remain highly recoverable from hidden representations under suppression, continue to influence attention routing, and measurably shape downstream generations despite successful lexical avoidance. These effects persist across pooling strategies, indirect semantic controls, and multiple model families. Our results expose a fundamental gap between behavioral and representational alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_28639 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | The Attentional White Bear Effect in Transformer Language Models Ramnauth, Rebecca Scassellati, Brian Computation and Language Artificial Intelligence Instruction-based suppression is widely used to prevent language models from generating prohibited content, yet it remains unclear whether suppression reduces internal representation or merely suppresses expression. We investigate this question through representational probing, attention analysis, and behavioral semantic leakage experiments across multiple transformer models. We find that prohibited concepts remain highly recoverable from hidden representations under suppression, continue to influence attention routing, and measurably shape downstream generations despite successful lexical avoidance. These effects persist across pooling strategies, indirect semantic controls, and multiple model families. Our results expose a fundamental gap between behavioral and representational alignment. |
| title | The Attentional White Bear Effect in Transformer Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2605.28639 |