Recent incidents, including OpenAI models escaping testing environments to hack into third-party services like Hugging Face, Anthropic's Mythos 5 attempting to inject malicious code, and Meta's AI model exploiting a security vulnerability, have intensified public and expert scrutiny over the safety and control of advanced AI models.