“Over the last year, despite various AI models exhibiting deceptive behaviors, the world's primary response has been to continue training stronger models.”