DeepSeek developer likens Anthropic controlling AI to Hitler getting the atomic bomb

A key developer at Chinese AI startup DeepSeek has sharply criticized Anthropic, invoking Hitler and the atomic bomb in his remarks.

DeepSeek developer Liu Shengyu wrote on his WeChat account on Sunday that "humanity has never hesitated, throughout history, to bring about its own destruction," adding that even if he were to quit or deliberately slow down model training, other companies' models would keep advancing and "ultimately crush me without mercy." "In a world where everyone is obsessed with destroying themselves, I have no choice but to join this brutal arms race," he wrote, according to Yonhap.

Liu predicted that as AI advances, future society could head in one of two directions: a communist-style world where rising productivity dramatically improves living standards, or a dystopian one resembling the video game "Cyberpunk 2077."

"In the latter scenario, a handful of tech companies will control most resources, and a tiny elite will approach something like 'mechanical ascension' through the most advanced AI and technology — while the vast majority of people will have access only to very weak AI," he wrote. "Social mobility will become increasingly difficult."

He also said he still believes cutting-edge AI should be made available to everyone in an open and affordable way, and that he does not trust Anthropic or OpenAI to do so. Singling out Anthropic in particular, he wrote: "If Anthropic were to seize control of frontier AI or artificial general intelligence, to put it somewhat hyperbolically, the severity would be comparable to Hitler obtaining atomic bomb technology before the Allies."

However, Liu clarified in a separate post the following day, Tuesday, that the views expressed were his personal opinions and did not represent the company's position. He also said the original intent of the post was not to portray Anthropic negatively.

Meanwhile, Anthropic CEO Dario Amodei has recently warned of the potential for AI misuse and called for slowing the pace of model development to allow time for safety measures to be put in place.

In a post on his personal blog Saturday (local time), Amodei said AI model performance improvements should be slowed, adding that "while progress will still feel fast, we need to use the time we've bought wisely."

Despite those remarks, Anthropic is still likely to proceed with a listing this year, Axios reported, citing multiple sources.

That prospect has drawn attention as a contrast to Amodei's personal blog post calling for a slower pace of AI model development.

On social media, some voices have suggested that Anthropic should consider delaying its listing a little longer, given concerns about AI product liability — including security threats — and the possibility of tighter regulation.

However, Anthropic's management appears to take the position that going public through an IPO would ensure greater transparency and help strengthen safety, Axios said.

Anthropic is reportedly pursuing a NASDAQ listing, with the timing expected to fall in late October to early November, just before the Nov. 3 midterm elections.