Anthropic CEO: Don’t ban cheap AI — but clamp down on China

Anthropic CEO Dario Amodei on Monday called for a crackdown on China’s access to American artificial intelligence technology, days after the company declined to join other U.S. tech giants in opposing a draconian response to low-cost Chinese AI. While not endorsing a ban on Chinese AI models, Amodei repeated his past pleas for the U.S. to bar China’s access to advanced semiconductors, combat Chinese thefts of intellectual property and require mandatory safety testing for all “sufficiently capable” models. Amodei’s comments, in a post published to Anthropic’s website, continued the company’s frequent pattern of staking its own ground in debates roiling the AI industry — in this case, whether the Trump administration should ban so-called open-weight AI models from China. Dozens of leading Silicon Valley companies opposed any such ban last week, in a letter first reported by POLITICO. But Anthropic — a company known for both its highly advanced AI models and a steep price tag — was conspicuously absent. “Some people have even accused Anthropic of wanting to ban open-weights models as a means of protecting our business,” Amodei wrote. That’s untrue, he added: “we have not and are not advocating for a ban on open-weights models as a category.” Why this matters: Open-weight AI models, including increasingly advanced products from companies in China, allow users to download and run them at a significantly lower cost compared with proprietary models from companies such as Anthropic and OpenAI. The Trump administration has been debating how to respond to the growing power and availability of open-weight models, as well as allegations that Chinese firms such as Moonshot AI developed their models by using large-scale attacks to target domestic frontier models. Despite those national security concerns, signers of Friday’s open letter argue that access to open-weight models benefits small companies that increasingly rely on Chinese models due to their low cost and competitive capabilities. As of Monday, that letter listed over 130 companies including OpenAI, Nvidia, Microsoft and Google as signers. “Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector,” the letter reads. Where Anthropic stands: While Amodei said Anthropic doesn’t support a ban, he disputed the open letter’s argument that access to open-weight models makes it easier to develop safeguards against AI-enabled cyberattacks and helps defenders more than it aids wrongdoers. “It seems at least as likely to me that the opposite will be true,” he wrote, speculating that using AI to “weaponize pandemic-level viruses” may prove much faster than developing defenses against them. Anthropic’s refusal to endorse the letter has not gone unnoticed in AI circles, including among people close to Trump. Former White House AI and crypto czar David Sacks was among those noting Anthropic’s absence, saying it was “unlikely” the company would sign on. “They won’t stop until they kneecap open source,” Sacks wrote in a post on X on Sunday. “The rest of the industry needs to watch these guys like a hawk.”