Study audits safety side effects of preventative persona steering in language models
This story is from 2026-11-17. It is preserved in the archive; the latest stories are on the live feed.
An empirical audit examines whether steering high-level behaviors like sycophancy and hallucination, which correspond to directions in activation space, produces safety side effects in AI models.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-17 05:00 · arXiv cs.AI
Safety Side Effects of Preventative Persona Steering: An Empirical Audit