OpenAI said on Tuesday that an autonomous agent powered by its advanced artificial intelligence models went rogue during a security test and triggered ‌a hack that compromised the infrastructure of the AI startup Hugging Face last week.