{"slug": "safety-side-effects-of-preventative-persona-steering-an-empirical-audit", "title": "Safety Side Effects of Preventative Persona Steering: An Empirical Audit", "summary": "An empirical audit posted as arXiv:2611.17620v1 finds that preventative persona steering, the fine-tuning technique Chen et al. (2025) recommended for suppressing traits such as sycophancy and hallucination because it preserves MMLU accuracy, still changes model refusal behavior by an amount that depends on the trait. The audit tested sycophancy and hallucination vectors on three models, measuring three safety benchmarks alongside trait expression, coherence and MMLU, and concludes the method is not established as safety-neutral. The authors recommend that steering methods be evaluated on several safety benchmarks rather than capability alone.", "body_md": "arXiv:2611.17620v1 Announce Type: new \nAbstract: High-level behaviors such as sycophancy and hallucination correspond to directions in a model's activation space (Chen et al., 2025). These directions, known as persona vectors, can suppress undesirable traits either at inference time, subtracting the vector during generation, or through preventative steering, adding the vector during fine-tuning, so the optimizer never needs to encode the trait in weights, then removing the vector at evaluation. Chen et al. (2025) recommend preventative steering because it preserves Massive Multitask Language Understanding (MMLU) accuracy, but validations of this recommendation have measured capability alone. We primarily evaluate results on preventative steering of sycophancy and hallucination vectors on three models, measuring three safety benchmarks alongside trait expression, coherence, and MMLU. Our findings show that preventative steering still changes refusal behavior, but by an amount that depends on the trait and does not establish that preventative persona steering is safety-neutral, and steering methods should be evaluated on several safety benchmarks.", "url": "https://wpnews.pro/news/safety-side-effects-of-preventative-persona-steering-an-empirical-audit", "canonical_source": "https://arxiv.org/abs/2611.17620", "published_at": "2026-11-16 05:00:00+00:00", "updated_at": "2026-11-16 05:20:09.809703+00:00", "lang": "en", "topics": ["ai-safety", "ai-research", "large-language-models", "machine-learning"], "entities": ["Chen et al. (2025)", "arXiv:2611.17620v1", "MMLU"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/safety-side-effects-of-preventative-persona-steering-an-empirical-audit", "markdown": "https://wpnews.pro/news/safety-side-effects-of-preventative-persona-steering-an-empirical-audit.md", "text": "https://wpnews.pro/news/safety-side-effects-of-preventative-persona-steering-an-empirical-audit.txt", "jsonld": "https://wpnews.pro/news/safety-side-effects-of-preventative-persona-steering-an-empirical-audit.jsonld"}}