AI Safety Researchers Find Flaws in OpenAI’s Internal Monitoring System Program Business