The unveiling of Moonshot AI’s Kimi K3 model at the World Artificial Intelligence Conference in Shanghai on 17 July sent a shock through US technology markets. Over the following four trading days, implied valuations for Anthropic and OpenAI on IG’s retail platform slumped by $232bn and $160bn respectively, a rout that analysts likened to the “DeepSeek moment” of the previous year. The 2.8-trillion-parameter open-weight model, which Moonshot claims rivals the performance of leading American systems, strained the start-up’s own capacity as demand surged, forcing a temporary pause in new subscriptions.

Viewed from Washington, the launch triggered an immediate escalation. White House science and technology adviser Michael Kratsios stated that the US government possessed information showing Moonshot had “distilled” Anthropic’s most capable model, Claude Fable 5, to build K3. He described a sophisticated internal platform designed to switch access methods rapidly and evade detection, and separately alleged that Moonshot acquired servers equipped with Nvidia’s restricted GB300 chips, accessing them via Thailand. Treasury Secretary Scott Bessent warned that sanctions “will be on the table” for industrial-scale intellectual-property theft, adding that US authorities were finding watermarks of American large language models on Chinese systems.

The accusations land as both Anthropic and OpenAI are manoeuvring to shape regulatory frameworks far beyond the United States. In Australia, the two companies welcomed Canberra’s plan to introduce direct AI regulation, a stance that equity analysts say provides clarity for investors and helps justify high valuations ahead of Anthropic’s planned public listing. Australia’s assistant technology minister framed the rules as a way to ensure creatives are paid when their work is used for training, while also signalling that clear standards could enhance the country’s attractiveness to big tech. Anthropic’s chief executive has separately called for a coalition of democracies to dominate AI and lock out China, a position that merges national-security arguments with commercial self-interest.

Moonshot has said it will release Kimi K3 as a fully open-source model on 27 July, a move that would allow developers worldwide to download and modify it. Beijing has not yet responded to the latest US allegations, though it has previously insisted that Chinese AI advances are the result of indigenous effort and international cooperation. The US State Department had already ordered a global diplomatic campaign in April to highlight what it calls widespread extraction of American AI intellectual property by Chinese firms. The next concrete milestone is the open-source release, which will test both the model’s real-world performance and Washington’s willingness to impose penalties on a company that has rapidly narrowed the gap with the West’s most advanced systems.