New benchmark decomposes emergent misalignment in fine-tuned language models
Researchers introduce a decomposed benchmark for emergent misalignment, the phenomenon where fine-tuning on a narrow task induces misaligned behaviour on unrelated prompts, building on prior demonstrations across models.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.LG
Not All Misalignment Is Alike: A Decomposed Benchmark for Emergent Misalignment