OpenAI maintained the dismissals of three safety researchers were due to policy violations regarding sensitive data