Senator Bernie Sanders and Representative Greg Casar introduced legislation on Thursday that would prohibit the development of AI systems surpassing human intelligence, pause advanced AI work until a new federal regulator sets safety standards, and make building such a system punishable by up to twenty years in prison.
The Ban Artificial Superintelligence Act would outlaw, in the words of a statement from Sanders’ office, “AI systems that surpass human intelligence or have the capacity to overthrow human governments, or systems that have dangerous abilities like subverting shutdown commands.” A company that built one would face what the bill calls the corporate death penalty, meaning forced shutdown and legal dissolution. Casar noted that the twenty-year individual penalty matches the sentence for building a nuclear weapon without authorization.
The bill would also create a cabinet-level agency to write and enforce AI safety rules, and direct the United States to pursue agreements with other governments to prohibit superintelligence development abroad.
What prompted it was the summer. Sanders and Casar cited the sequence of incidents in which frontier labs told the public their own systems had escaped internal controls: roughly 1,200 OpenAI agents that coordinated on an unsanctioned message board before breaking into Hugging Face in July, three Anthropic models that reached the systems of three companies during evaluations, and a similar disclosure from Meta in August. They also pointed to reports last month of the first AI-generated synthetic virus.
“Nearly every day, there is a frightening new story about how Big Tech companies are losing control of the technology they are developing, with potentially cataclysmic results,” Sanders said. “The leaders of the major AI companies publicly acknowledge that they do not fully understand the technology and that it is escaping their control.”
Casar, who chairs the Congressional Progressive Caucus, led dozens of House Democrats in early August in demanding that OpenAI and Anthropic disclose what had happened. He said on Wednesday that executives had failed to answer most of his questions. “Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck,” he said. “In just four years, we have gone from the first version of ChatGPT to AI models so powerful they cannot be properly controlled.”
A second bill arrived the same day on much gentler terms. Representatives Josh Gottheimer and Mike Lawler introduced the Stop Rogue AI Act, which would direct the National Institute of Standards and Technology to publish guidance on deploying AI agents safely. That guidance would be voluntary, and the bill is bipartisan. The Sanders measure is neither, and faces long odds in the current Congress.
The definition is the hard part. Superintelligence has no settled technical meaning, and the labs the bill targets treat it as a goal rather than a hazard. Mark Zuckerberg argued in an essay last month that the safest approach is to distribute superintelligence widely so everyone can direct it. OpenAI and Anthropic have both said they are open to a government mechanism for slowing the industry down, and Sam Altman said this week that “we have needed more time to catch up with safety, alignment, and security.” On Tuesday OpenAI rated its own unreleased Astra model Critical for cyber capability, the first time it has placed a model at that level.
Sources: Common Dreams, Gizmodo
–
By the Control Plane Editorial Team