Review
Open access
Sep 2026
AI safety is a system property, not a sum of individual defenses
Summary Safety for large language models (LLMs) and artificial intelligence (AI) agents remains fragmented; defenses against jailbreaking coexist with unresolved hallucination, fairness, and autonomous-control problems. Existing frameworks address these risks, but few specify how failure in one dimension changes anothe...
Xiang Chen, ,-Chen-Xiao Lin, Long-Long Zhu et al.
· Patterns · 0 citations