AI according to Anthropic: 'If China recovers, global cybersecurity is at risk'
Dario Amodei's company calls for a halt to semiconductor smuggling to Beijing and denounces the risk of a global cybersecurity dominated by authoritarian regimes
Stop the illegal import of chips into China and the distillation of the responses of American models for training Chinese ones. While US President Donald Trump is visiting Beijing, Anthropic makes this appeal. And it does so through a paper in which the competition in the field of artificial intelligence becomes the epochal struggle between democracy and authoritarian regimes.
Yes, because the report '2028: Two scenarios for global AI leadership', which appeared on the blog of the company that develops Claude, paints precisely two scenarios. One in which Washington maintains a 12-24 month competitive advantage over its Asian competitors, maintaining its leading role in the development of a technology that 'protects civil liberties'. The other, at the opposite end of the spectrum, in which China catches up on much of the technology gap that exists today, AI is rapidly adopted by both private companies and the state, and the Chinese Communist Party (to which, it must be said, the report often refers when speaking of China) becomes a threat to global cybersecurity.
It is no coincidence that the alarm clock, for Anthropic, sounded thanks to Claude Mythos, the model made available only to certain companies as part of Project Glasswing. A version so good at finding flaws in the code that using it Mozilla, the foundation that develops the Firefox browser, found more bugs in April alone than it had found in the previous 15 months. And two out of three flaws posed high security threats. On his blog, however, the research director of the cybersecurity company Crowdfense, Paolo Stagno, downplayed these numbers, pointing out that not all bugs can be exploited by a malicious hacker.
However, it is not only a question of cyber security. In the report, Anthropic points out how AI is a dual technology, i.e. it also has applications in the military. "The Chinese Communist Party is building its army for a battlefield integrated with artificial intelligence," Anthropic writes.
To avert this, argues Dario Amodei's company, the same one that entered into controversy with the Pentagon to which it refused to supply an AI to be used for defence purposes without any kind of restriction, it is necessary not only to maintain the export ban on chips to China. But also stop the smuggling of them. And put a stop to the distillation of American models, which Anthropic itself denounced a few months ago as common practice by Chinese artificial intelligence companies.
