‘AI going off the rails’: Techbro buys island to experiment with AI government, as sign-ups soar, he worries it won’t end well

Last year, tech founder Dan Thomson announced his plan to launch an AI-governed country. He even claimed to have acquired an island in the Philippines, naming it after his company, Sensay. Now the project is moving forward with thousands of sign-ups, even as its creator admits he is not entirely sure how it will end. 

Per CNN, the island is intended to function as a micronation, governed by a council of AI-powered bots modeled after historical figures like Winston Churchill, Eleanor Roosevelt, and Marcus Aurelius. While the island currently has no official international recognition, Thomson has already opened applications for residency, and he reports that around 12,000 people have already registered to be residents. 

The governance structure involves residents proposing ideas, which the AI council then reviews and votes on. From there, humans are expected to carry out the AI’s decisions, although finding people willing to follow these decisions to the letter is a challenge. He believes the process will become more automated as the technology evolves. He envisions a future where AI can hold its own crypto wallets and bank cards to hire contractors and manage manual labor independently.

Most AI utopias in fiction fail. Then again, usually because of humans

Some applicants are drawn to the project because they are tired of traditional politics. Piotr Pietruszewski-Gil, a project manager for the island, noted that many are fed up with corruption and broken promises. 

Thomson shares this sentiment, suggesting that AI leaders could offer a more objective approach to governance. Essentially free from the personal motivations and lobbyists that often influence human politicians.

Alondra Nelson from the University of Oxford’s Institute for Ethics in AI, called the idea ridiculous. She noted, “I mean, we see AI literally going off the rails every single day, right? If you think about the example of the ‘nudified’ feature on Grok, you think about the examples of the young people who are committing suicide based on their engagement with these technologies.” 

Consider another recent, grievous error: naming a school as a military target in Iran. Doesn’t her concern seem valid? Nelson also argued that the project’s structure is inherently anti-democratic because it is controlled by a single founder.

Thomson himself has expressed some uncertainty about the outcome of his experiment. When asked about the potential for the AI to behave aggressively, he said, “Um, if it starts acquiring weapons and attacking neighboring islands, that would be a bad situation,” before adding, “I think it’s extremely unlikely.” 

He also remains open to the possibility of the project taking an unexpected turn, stating, “If we literally end up with Stalin running a government, on top of Mussolini, Hitler, and, you know, Genghis Khan — you know, that is the social experiment, and that’s the way it will end up.”

For now, the project continues to move toward its 2027 residency launch. Thomson remains fascinated by the potential risks and outcomes, admitting that the biggest danger might actually be the human influence on the AI’s development. 

He stated, “I think the biggest risk of that is having someone suggest that they should be — you know — fierce, and powerful, and conquering, and having a Caesar-type character coming in to try and capture more land. Honestly, I wouldn’t know. I’m fascinated to find out.”