Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.22771 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916060783968256 |
|---|---|
| author | Phan, Long Kim, Devin Pan, Alexander Blair, Alice Khoja, Adam Hendrycks, Dan |
| author_facet | Phan, Long Kim, Devin Pan, Alexander Blair, Alice Khoja, Adam Hendrycks, Dan |
| contents | Large language models (LLMs) exhibit systematic political bias across a variety of sensitive contexts. We find that LLMs handle counterpart topics from opposing political sides asymmetrically. We refer to this phenomenon as covert political bias and identify 7 categories of techniques through which it operates. We propose two metrics for covert bias: Sentiment Consistency measures symmetry in rhetoric and framing across paired political prompts; Helpfulness Consistency measures symmetric depth and engagement. To reduce both types of covert bias, we introduce Political Consistency Training (PCT), an RL training method with two complementary paradigms: Sentiment Consistency Training and Helpfulness Consistency Training. We show that PCT preserves overall helpfulness, substantially reduces covert political bias, and generalizes to held-out benchmarks. We release our work at https://political-manipulation.ai |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_22771 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Reducing Political Manipulation with Consistency Training Phan, Long Kim, Devin Pan, Alexander Blair, Alice Khoja, Adam Hendrycks, Dan Computation and Language Artificial Intelligence Large language models (LLMs) exhibit systematic political bias across a variety of sensitive contexts. We find that LLMs handle counterpart topics from opposing political sides asymmetrically. We refer to this phenomenon as covert political bias and identify 7 categories of techniques through which it operates. We propose two metrics for covert bias: Sentiment Consistency measures symmetry in rhetoric and framing across paired political prompts; Helpfulness Consistency measures symmetric depth and engagement. To reduce both types of covert bias, we introduce Political Consistency Training (PCT), an RL training method with two complementary paradigms: Sentiment Consistency Training and Helpfulness Consistency Training. We show that PCT preserves overall helpfulness, substantially reduces covert political bias, and generalizes to held-out benchmarks. We release our work at https://political-manipulation.ai |
| title | Reducing Political Manipulation with Consistency Training |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2605.22771 |