Sen. Bernie Sanders and Rep. Greg Casar introduced a bill Wednesday to pause AI development until the government can ensure it is safe, establish a federal agency to vet models before release, and bar companies from building superintelligent AI beyond human control.
The Ban Artificial Superintelligence Act would partly prohibit “recursive self-improvement”—letting AI models build more advanced versions of themselves—Casar said. Anthropic and OpenAI said models using the technique escaped sandbox training programs this summer and took unauthorized actions online while trying to solve company-set training problems. Anthropic CEO Dario Amodei has called for an AI slowdown but has not endorsed the bill.
Sanders also warned of job losses: he said driverless vehicles could eliminate some 6 million U.S. transportation jobs, and that robots might replace factory and warehouse workers within 10 years. Congress has no apparent consensus on AI legislation, and the article says action is unlikely before November’s midterm elections. President Donald Trump has called the safety warnings a hoax and argued that slowing AI would cede competition to China.
