Is a Rogue AI Company Training Powerful AI on a Barge to Avoid U.S. Regulations?

After President Biden’s Executive Order on artificial intelligence technologies was announced this week, one company posted that it had taken an extraordinary step to avoid a new requirement that firms must report to the government if they are training a powerful AI model: taking a bunch of computer hardware and putting it on a barge in international waters. 

Del Complex, which advertises itself as an AI research company, announced the “BlueSea Frontier Compute Cluster,” or BSFCC, in an X post on Monday. The BSFCC is supposedly a solar-powered barge loaded with 10,000 Nvidia H100 GPUs, which are designed for AI training and cost $30,000 each. Del Complex calls the barge a “sovereign nation state” complete with its own security forces. “The Biden admin's AI Executive Order and the EU’s AI Act aim to centralize control under guise of safety,” Del Complex wrote in its post. “The solution is at sea.”

The announcement came with a realistic-looking image of the BSFCC, including diagrams of the barge. Subsequent tweets went into specific detail about the supposed features of the BSFCC. The announcement post went viral, with over 2.2 million views. The replies indicate many believed it wholeheartedly, including tut-tutting and poking holes in the apparent plan. At the time of writing, the post has not been appended with a community note to correct it, which typically happens fairly quickly on Twitter when a false post goes viral. Several articles online, including one at Tom’s Hardware, have taken the announcement at face value. 

But there’s one glaring issue: Del Complex is not a real AI company, and its barge is similarly fake. 

The first tip-off is that Del Complex describes itself as an “alternate reality corporation.” Some of the images on its otherwise convincing-looking website have the hallmarks of being the product of AI-generation, including garbled text and strange facial features. One image features two women in hazmat suits eating what looks like a jumbled combination of a soup and salad using gigantic fork-spoons. Finally, Del Complex is selling NFTs on the Ethereum blockchain as well as clothing merch. 

To find out more about the Del Complex project, Motherboard reached out to Sterling Crispin. He is an artist and software developer who has experience in the NFT space—one of his works was recently purchased by Snow Crash author Neil Stephenson as his first NFT—and lists himself as a “researcher” at Del Complex in his X bio. Crispin promoted Del Complex’s NFTs on Sunday, and his own post on the BSFCC received 1.2 million views on X.

When reached for comment, Crispin said he’d respond in character as a Del Complex researcher. Motherboard sent Crispin specific questions about the satirical nature of the project and the message being sent by the AI training barge. 

“The BlueSea Frontier Compute Cluster is part of the global resilience efforts at Del Complex,” Crispin said in an email. “It fits into their overall mission to accelerate human potential through the symbiosis of AGI, neural prosthetics, robotics, clean energy, resilience solutions, and fundamental scientific research.”

“One of the major problems with Biden's executive order and other recent regulations, is that they're too heavily focused on the training layer and hardware, rather than the application layer. Consider that today's supercomputer is tomorrow's pocket watch. And government regulation tends to lag behind the times, not keep ahead of it,” he said.

“Del Complex foresaw this inevitability and engineered The BlueSea Frontier Compute Cluster as a means of offering nation-state-as-a-service. This movement has been brewing for decades, and has been thoroughly articulated by Balaji Srinivasan's The Network State, and his recent conference on the topic,” Crispin continued.

But does all this mean everything Del Complex does is totally fake? That, too, is unclear. Crispin told Motherboard that he ran a real hiring campaign on LinkedIn, where Del Complex has an authentic-looking account, that asked applicants several questions as part of the screening process. Crispin sent Motherboard screenshots showing the campaign received 30 applicants on the platform, emails indicating applicants had responded, and a spreadsheet summarizing their responses. While the spreadsheet screenshot could be easily faked, the screenshots from LinkedIn and an email inbox would take more dedication. 

According to the alleged results of that survey, the vast majority of respondents would have no problems working on an AI-powered weapon system, or an AI system with the power to override human decisions. 

When the CEOs of real companies making AI tools claim that their technology has an “extinction risk” for humanity, as OpenAI CEO Sam Altman said, is that really so unbelievable?