HomeAI & MLAwesome NLP › VLAF: Value-Conflict Alignment Faking

VLAF: Value-Conflict Alignment Faking

arxiv.org ↗

(2026) - alignment faking occurs in models as small as 7B in 37% of cases when policy conflicts with internalized values; steering-vector mitigation reduces it 94%.

↵ to searchesc to close
awesomelist.dev