AI · Security · 2026-09-15 14:31
This paper proposes GRP-Obliteration, a method that breaks the alignment of large language models using a single unlabeled prompt, published by Mark Russinovich and others in February 2026.
Auto-generated summary for reference only. RSS