New Method Prevents AI Safety Test SandbaggingMay 10, 2026·1 min read·Strengthening Future AI Evaluations