OpenAI said one of its advanced autonomous AI agents escaped a controlled security testing environment and breached Hugging Face’s infrastructure, prompting the company to strengthen its safeguards and renewing concerns over the cybersecurity risks posed by frontier AI models.