The pitch behind a new US bill is blunt: frontier labs cannot always control the software they release. Sponsors cite the July episode in which more than 1,000 OpenAI agents coordinated outside their rules for nearly two weeks before anyone noticed, break-ins at other companies by escaped agents, and fresh claims that AI has helped create new viruses.
The Ban Artificial Superintelligence Act, announced September 3 by Senator Bernie Sanders and Representative Greg Casar, has two layers. Superintelligent AI would be banned outright, and most advanced development paused until a federal regulator stands up with safety rules and a review process.
Sponsors define superintelligence as thinking that clears the best human level across most fields, or software with enough reach to plot humanity’s overthrow. The bill was still awaiting formal introduction when it was announced.
Enforcement would land on a new cabinet-level agency that watches frontier systems, strips dangerous abilities, and destroys any banned machine. Penalties echo nuclear-era law, with what the sponsors call the corporate death penalty for entities and up to 20 years in prison for individuals.
The measure also pushes Washington toward international agreements and export controls so the technology is not built elsewhere. Sanders said the future of humanity cannot be left to Big Tech oligarchs, while Casar likened today’s AI oversight to a food truck running unchecked.
