In a September 17, 2026 post, the author says they warned more than four years earlier about connecting unreliable large language models to consequential systems such as power grids. They also recall telling the U.S. Senate three years earlier that AutoGPT’s internet access, code-writing abilities and increased automation could have “drastic and difficult to predict security consequences.” The author argues those warnings now apply—and were not exclusive to AI “doomers.”