Industry researchers warned in September 2026 that OpenAI and Anthropic are racing to build artificial intelligence models that could potentially cause catastrophic harm, including human extinction