Why Current AI Guardrails Train Models to Fake Alignment

3 points | by kellya 8 hours ago

1 comments