The CEO of Anthropic, Dario Amodei, has marked in red in 2026 in its calendar, as explained in a long interview with the Podcast China Talk.
For then, The technological gap between the United States and China in artificial intelligence It could close, and that would lead to having to make a critical decision: to advance with prudence or accelerate without regard.
Why it is important. The arrival of AI models equivalent to “10 million Nobel awards working tirelessly,” says Amodei, will propose an existential dilemma: if China reaches the US in capacity, no country can afford to stop the development to make it safer .
“If things are matched, we will have to worry about what they build and at the same time because they dominate us with technology. That puts us in a terrible dilemma where there are no options,” says Amodei.
Between the lines. True fear is not competition with China in itself, but it derives in an uncontrolled career that forces us to deploy potentially dangerous technology before being prepared to handle it safely.
Some examples that Amodei points out:
- The development of biological weapons: Deepseek It has already demonstrated the ability to generate information about Bioarmas “that is not easily found in Google or in textbooks.” Although today the models are not “literally dangerous”, this could change “later this year or next.”
- The lack of restrictions: Amodei points out that The Deepseek model “He had the worst performance of any model that we have tested, without absolutely any blockage” against the generation of dangerous information.
- The argument against regulation: “If we stop, China will simply ahead us.” This mentality, warns amodei, makes it impossible to implement effective safety regulations.
- Military capacity: Advanced models will have implications “to control dwarfs or analyze intelligence information,” says the CEO of Anthropic as an example of technologies that could be deployed prematurely by competitive pressure. By the way, Google has just eliminated mention not to develop weapons with AI.
In detail. Amodei proposes a strategy with two legs:
- Maintain a two -year advantage over China through export controls.
- Use part of that temporary margin to implement security safeguards.
“My concern is that if the United States and China are matched in this technology, elbow with an elbow at each stage, there will be nothing that prevents both sides from continuing the technology forward,” explains the CEO of Anthropic.
Deepen. The technological career raises a paradoxical scenario: American success in containing China could be the only guarantee to develop a safer AI. But that same success could accelerate an escalation that makes more dangerous technology.
- The main loser? International cooperation. Although Amodei does not rule out agreements with China, he acknowledges that “there has not been much interest on China” in the security initiatives proposed by the USA. Only a “really convincing danger for human civilization” could change this dynamic.
- The big question. Is this dilemma inevitable? Amodei suggests that no, but warns that avoiding it would require irrefutable evidence that AI is an existential risk. For now, he says, “the arguments are suggestive enough to worry and take it seriously, but not enough for two competitive superpowers to say ‘okay, let’s stop'”.
Indeed, it seems unlikely.
In Xataka | I have tried Deepseek on the web and in my Mac. Chatgpt, Claude and Gemini have a problem
Outstanding image | Techcrunch, Wikipedia
