Commentary: Anthropic’s retreat is a win for Chinese open-source AI

Commentary: Anthropic’s retreat is a win for Chinese open-source AI Cheaper AI was already a major selling point for Chinese labs, and now they can add that they’re more reliable, says Catherine Thorbecke for Bloomberg Opinion. TOKYO: Ahead of its launch, Anthropic's most powerful artificial intelligence model, Mythos, was characterised by some in China as a “reverse DeepSeek” moment. It was a reminder that for all the hype around Chinese challengers, America’s leading AI firms were still on top. Yet somehow the botched rollout of Mythos and its defanged sibling Fable has turned into an incredible free advertisement for Chinese open-source AI. Shortly after Anthropic announced that it was disabling access to Mythos and Fable after the Trump administration ordered it to bar the tools from foreign nationals, Chinese AI lab Zhipu jumped at the opening. The Beijing-based firm said it was launching its most advanced model yet - and that it was making it open source. The sudden restriction of certain models is “deeply regrettable”, Zhipu co-founder Jie Tang wrote in a post on X. “At a time when access to frontier models is abruptly cut off for non-technical reasons, we are even more convinced of one thing: Science should be global.” Tang said their new model would launch to certain users at 5:21 - seemingly taking aim at Anthropic’s own memo stating it received Washington’s directive at 5:21 pm. Making the model open means that companies, governments or organisations with sufficient hardware can run it locally, and never have to worry about it being yanked on a whim. Shares for Zhipu, which trades as Knowledge Atlas Technology JSC in Hong Kong, surged on Monday. As one analyst told Bloomberg News, the firm was sending a powerful message at a time when Anthropic is being forced to restrict access. FRAGILITY OF AMERICAN AI DOMINANCE Instead of showing America’s AI dominance, as was intended, this debacle just exposed its fragility. As I’ve written before, the performance gap between the top-of-the-line Chinese AI models and the best Silicon Valley offerings this year was widening, not shrinking. But all that means nothing if the world can’t access the technology. As Kyle Chan of the Brookings Institution wrote, some open-source Chinese AI firms are having a “field day”, being handed a free sales pitch they couldn’t even have fathomed. It doesn’t help that these export controls didn’t just target adversaries, but American allies and even foreign members of Anthropic’s own staff - the very people who helped build the models. Washington’s ambitions for the world to build on the American AI stack look much weaker if that also comes with uncertain politics. Other nations are taking notice. A French presidential candidate called the move a wake-up call, adding that “a nation that depends on others for its technology is a nation that can be unplugged overnight”. Canadian Prime Minister Mark Carney declared: “We will have done something wrong if we just accept this, don’t take this lesson, don’t build out and diversify.” Global leaders are naming the very fears that Chinese AI firms are now trying to monetise. Bringing Mythos and Fable back online overnight is unlikely to silence this growing chorus. THE BEGINNING OF A PRICE WAR The timing, meanwhile, couldn’t be worse. This is all happening in the midst of a brutal debate over AI cost and the beginning of a price war. As astronomical AI spending bills are starting to come due, more business leaders are being forced to reckon with whether the top models are worth it. One of the biggest draws of China’s AI offerings is that they are just a fraction of the top models from Anthropic and OpenAI. More people seem to be switching over: DeepSeek, which last month said that it was making permanent a 75 per cent discount on its latest model, remains at the top of OpenRouter’s ranking of large language models by token usage (the units of data processed by an AI tool). US AI still tops the elite benchmarks, but most companies don’t need systems that can reason through quantum physics to automate routine desk work. Good-enough and cheaper AI was already a major selling point for Chinese labs, and now they can add that they’re also looking more reliable. Some outlets have reported that the Mythos and Fable crackdown was in part spurred by fears that it could be accessed by China (though Anthropic and former White House AI czar David Sacks have only said that it was due to potential jailbreak risks). Yet if Washington’s goal was to protect America’s AI lead with this move, it has backfired. The whole thing is incredibly awkward for Anthropic, a company that has long cast itself as the safety-focused AI lab that welcomes more regulation. Now that it is caught in the middle of a chaotic regulatory intervention, supposedly over risks, it’s insisting that its guardrails are enough. But this is about far more than Anthropic. Ultimately, Chinese AI labs are right to welcome this move, but framing this as an easy win for Beijing misses the larger failure. The rest of the world should be more worried than amused. The US-China competition lens has flattened all serious debates about AI safety. The result is that the technology keeps moving faster than anyone’s ability to govern it. In that race, the loser may not be Washington or Beijing but everyone else.