Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLM Paper • 2509.18058 • Published Sep 22 • 12