Findings from a Pilot Anthropic–OpenAI Alignment Evaluation Exercise