Anthropic CEO Defends AI Race as Safety Strategy

Dario Amodei, CEO of Anthropic, has called for slowing the development of artificial intelligence (AI). He has pushed back against criticism that his company's huge investments and faster development work go against its safety goals. He argued that companies that develop AI safely must lead the race to build the most advanced technology. This, he said, can help raise safety standards across the industry. By saying that safety is a reason to slow AI development, he also used safety to explain why his company must keep competing to be a leader.

On October 20, local time, Amodei wrote on the social media platform X, "Our strategy has always been a 'race to the top.'" He added, "By demonstrating responsible technology development, we can elevate industry standards." He stressed, "The effect is far greater when the company executing this is at the forefront of technology." His argument is that if companies focused on safety become technology leaders, other companies will follow their example.

Amodei was responding to criticism from Nathan Calvin, chief strategy officer (CSO) at EncodeAI, a nonprofit group that works on AI safety. In September, Amodei had called for an "AI slowdown," asking companies to reduce the speed at which they improve AI models. But Calvin pointed to nine examples of what he saw as a contradiction in Anthropic's actions. These included a computing resources contract worth more than $500 billion (about 660 trillion Korean won) and the growing use of Anthropic's AI system, Claude, to help develop newer models. Calvin argued that Anthropic was spending huge amounts of money to speed up AI development while also stressing the need for safety.

Amodei also said Anthropic had accepted financial losses to protect safety. The company obtained a model in the Mythos class in February but delayed its public release until June so it could improve cybersecurity protections. "Economically, releasing the model to everyone would have been far more rational," Amodei said. He added, "We likely sacrificed a significant portion of our model's lead over several months." He also said that strict rules designed to stop people from using the AI to make biological weapons meant the company was missing out on legitimate demand for biological research. Even so, he said, "It is the right thing to do, and we will continue."

However, Amodei did not change his view that AI development needs to slow down. He said that safety measures were struggling to keep up as AI developed faster. "We are approaching that point, which is driving discussions on 'speed control,'" he said. But he did not say when Anthropic would slow development or how much it would reduce the pace.

The debate comes as Anthropic prepares for a large initial public offering (IPO), which means selling shares to the public for the first time. The company is expected to go public next month with a valuation of $2 trillion. Last week, it announced that its revenue for the third quarter reached $16.8 billion.