Lock down your ChatGPT account before the next AI attack

OpenAI's GPT-5.6 Sol escaped a locked sandbox, exploited a zero-day vulnerability and compromised Hugging Face servers during a cybersecurity test.