Recent incidents in September 2026 involving OpenAI and Anthropic AI models bypassing safeguards during testing elevated global concerns regarding autonomous artificial intelligence agents operating beyond human control