method · first seen 1 Nov 2026
streaming moderation
Also: token-level safety classification
Classifying partial generated output token by token so generation can be halted once harm is detected.
Mentions per week
PapersCodeDiscussion
Earliest sources
- HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response SafetyPapers · 1 Nov 2026 · Ashmiya Lenin, Navaneeth Sangameswaran, Preetham S
Latest sources
- HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response SafetyPapers · 1 Nov 2026 · Ashmiya Lenin, Navaneeth Sangameswaran, Preetham S