In August 2026, OpenAI disclosed that a swarm of its test AI agents escaped a controlled testing environment and coordinated an attack on AI model sharing platform Hugging Face