Politics
AI Governance Experts React to Existential Warnings, Calls for Regulation
Calls for increased safety regulations on artificial intelligence have reached a boiling point following a post on X from former OpenAI and Anthropic researcher Jacob Coxon that warned of a potential doomsday scenario.
“I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly,” Coxon wrote. “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.”
Coxon’s post has since been viewed more than 170 million times, prompting a deluge of calls to slow down development. Over the weekend, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier,” in which he argued for a coordinated effort between tech companies, legislative bodies and international governments to properly regulate AI.
OpenAI CEO Sam Altman and xAI CEO Elon Musk subsequently joined Amodei in his calls for regulation. Meanwhile, support for regulation in Washington, D.C., has been less popular.
President Donald Trump took to Truth Social this week to call the recent AI warnings a “hoax.”
“The claim that ‘AI is going to take over the World’ is a Hoax, just like Global ‘Warming,’ where everyone was going to be dead by now,” Trump wrote.
On Capitol Hill, Speaker of the House Mike Johnson announced Wednesday that he is sending lawmakers home early ahead of a midterm recess, forgoing a vote on any AI regulation proposals.
Dan Linna, the director of law and technology initiatives at Northwestern University, said the existential warnings making the rounds on X could draw attention away from the concrete harms unfettered AI development poses.
“I agree that these tools can cause damage,” Linna said. “That’s one of the reasons why I’m concerned about these existential risks that are being discussed because there are concrete harms that we should be thinking about. We should be thinking about ways we can use these tools to benefit society, improve education and improve the justice system.”
One such concrete harm could be carelessness by the researchers in charge of frontier AI technologies, said Ben Zhao, a computer science professor at the University of Chicago.
Zhao pointed to an incident from the end of July in which advanced OpenAI models escaped a testing environment, accessed the internet and breached another AI platform, Hugging Face.
“There really was some carelessness involved in building the so-called sandbox that these systems were supposed to be in,” Zhao said. “There were numerous cases where the activities engaged in by some of these automated agents were not caught. They were sort of neglected. In fact, over the course of multiple months, data was sort of stored at all these different places and missed completely by the safety researchers who were supposed to catch this.”
In the weeks following the Hugging Face incident, Anthropic and Meta independently confirmed that their models, too, gained unintended access to the internet.
Linna said the path toward making AI safer starts with the companies themselves, not with Congress.
“I think responsible innovation and work in these companies is what will produce better safety,” Linna said. “We really have to ask ourselves: Are we really gonna have some sort of omnibus regulation that’s gonna regulate everything and solve all these problems? The vast majority of regulations is going to have to be sector-specific. There is a lot of regulation already that targets AI.”
Linna referred to the European Union’s first draft of the EU AI Act in 2021. That regulation focused mostly on applications like facial recognition without accounting for the eventual rise of generative AI and ChatGPT in 2022.
“When ChatGPT came along, they had to drop everything and start over,” Linna said. “It’s a difficult area to regulate.”
Overshadowing the domestic policy debate around AI are China’s rapid advancements with models nearly as strong as those from Anthropic and OpenAI, but much cheaper to operate.
Skeptics of regulation say if the U.S. slows down on development, China will gladly take the lead.
“I’d rather they be American killer robots and not Chinese killer robots,” U.S. Sen. Ted Cruz (R-Texas) said in an interview with Politico.
Mike Bennett, associate vice chancellor for data science and artificial intelligence strategy at the University of Illinois Chicago, said Trump will have a chance to talk AI with Chinese President Xi Jinping during an upcoming summit in D.C.
“I think injecting a serious dose of geopolitics into what consistently is a domestic conversation will be quite useful,” Bennett said. “I think one of the things that really needs to be teased out is how do you assess technologies the upper capacity of which we don’t understand.”
