In which year will the first major closed-source AI company experience a leak or theft of their frontier model weights? At present, I consider these the major US AI companies: OpenAI, Anthropic, and Google DeepMind. If other companies rise to the same level, I will include them as well. The main criteria I will look at when considering adding new companies are the benchmarks such as LMSYS Chatbot Arena leaderboard (typically top 10, excluding multiple models from the same company) and widespread usage (this can be seen through official release data or by comparing usage on OpenRouter to other major AI models). We are excluding companies that focus on open-weight models, such as Meta AI, due to there being less at stake when the primary goal is to open the weights at a later date. For this reason, I am hesitant to add xAI, as they have opened the weights to their first model, and I am uncertain about their future plans. For resolution of this I will look for news releases, Official releases, and other sources that can say with certainty that a leak or theft of model weights have occurred.