A small Israeli AI security startup has emerged as the common link in separate cybersecurity incidents involving OpenAI, Anthropic and Meta. But the story isn't simply about AI models escaping control ...