Findings from a pilot Anthropic–OpenAI alignment evaluation exercise: OpenAI Safety Tests – OpenAI