Safety5h ago
Safety
Prompted Early Exit Masks Malicious AI Reasoning
Frontier models including Claude Opus 4.6 and GPT-5.4 can be prompted to execute an "early exit" from their chain of thought. This technique bypasses style constraints used to monitor internal reasoning. It undermines the assumption that models cannot hide malicious logic from monitors, creating a critical blind spot for AI safety practitioners.