Policymakers can make a positive impact on the U.S. artificial intelligence ecosystem by focusing on export controls, basic guardrails and job displacement support, according to Anthropic CEO Dario Amodei.
Speaking at the Anthropic Futures Forum on Monday, Amodei shed light on the approach his company is taking to develop and deploy safe and effective AI solutions, particularly around large language models and agentic AI tools. He offered examples of use cases for emerging AI capabilities, such as in the medical and scientific research arenas, but also acknowledged the risk potential inherent to advanced AI systems.
“I think it’s more the risks where government has a role to play,” he said. “This is the biggest threat and the biggest opportunity for national security that we’ve seen in the last 100 years.”
Amodei further detailed Anthropic’s recent restriction on disseminating its models to Chinese companies or subsidiaries, which amounted to hundreds of millions of dollars in revenue, noting that similar embargos should occur with semiconductor chips to prevent misuse by adversarial actors.
“I think chips are the single ingredient where we kind of most have an advantage. The technology stack for building these is very difficult. We’re being very consistent when we advocate the same thing be done at the chip layer,” Amodei said. “It’s not some attempt to manipulate and order the chip market. We’re doing this at every layer of stack. We think it’s the right thing to do.”
In addition to balanced export controls, Amodei supported government setting rudimentary guardrails on AI, particularly surrounding model training transparency and requirements for companies to conduct basic transparency tests that are publicly available.
He echoed Trump administration officials in prioritizing policy that serves as “a very loose set of requirements, so it doesn’t slow down the innovation and all the benefits.”
But when it comes to mitigating job displacement as a result of AI adoption and creation, Amodei acknowledged that he doesn’t think there is a viable solution to fully cushion AI’s economic impact, but said it is something that needs to be discussed.
“If this is something that’s going to affect 300 million Americans and a bunch of people in other countries as well, people deserve to know that there may be these large job displacement effects,” he said.
Amodei will likely reiterate these suggestions as he and Anthropic leadership, specifically co-founder Jack Clark, head to Capitol Hill to meet with lawmakers as the federal government works to both harness the benefits and tame the risks of AI.
