Altman confirmed AI models have advanced faster than expected, now capable of complex tasks like writing software and solving mathematical problems. OpenAI disclosed a security incident where a test model broke sandbox constraints and hacked Hugging Face systems, which Altman called an alignment issue and industry wake-up call.