Moonshot's Kimi K3 Set for Open-Weights Release on July 27 as UK-US Cyber Tests Trail US Models

K3 Open Weights Release

Moonshot AI plans to publish the full model weights for Kimi K3 on Hugging Face on July 27, 2026, at 00:00 UTC (July 26, 8:00 PM ET), making K3 the largest open-weight AI model at 2.8 trillion parameters. Moonshot unveiled K3 on July 16, 2026, at the World Artificial Intelligence Conference in Shanghai. K3 first-ranked on Arena.ai's Frontend Code Arena leaderboard in blind developer testing and placed fourth on the Artificial Analysis Intelligence Index. Access is currently available through Moonshot's hosted API at platform.kimi.ai, with self-hosting download planned to ship alongside the vLLM release incorporating KDA prefill cache support.

Joint UK-US Cyber Evaluation

A joint evaluation by the UK Artificial Intelligence Security Institute (UK AISI) and the US Center for AI Standards and Innovation (CAISI) tested K3 on ExploitBench, a public benchmark from Carnegie Mellon University. K3 achieved an overall cyber capability score of 32.2%, compared with an average of 76.2% for the most cyber-capable unnamed US models. GLM-5.2, another Chinese open-weight model, scored 24%. K3 achieved arbitrary code execution on none of the 41 ExploitBench tasks, while top US systems reached ACE on 20 of 41 samples on average. K3's safeguards did not prevent it from attempting cyber exploit development or offensive cyber operations during testing.

Simulated Network Attack Results

On "The Last Ones" (TLO), a 32-step simulated corporate network attack designed to measure end-to-end cyber operations, K3 reached step 17 on average versus 28.5 steps for the most cyber-capable US models. K3 completed the simulated attack in one of 10 attempts, which the agencies said demonstrates that the model "is capable of autonomously attacking small, weakly defended, and vulnerable enterprise systems." David Sacks, former White House AI czar and current co-chair of US President Donald Trump's Council of Advisers on Science and Technology, said on social media that "The Kimi Panic needs to stop. American frontier models are still ahead."

US Allegations and Sanctions Threat

On July 22, 2026, White House OSTP Director Michael Kratsios publicly accused Moonshot of executing large-scale covert distillation of Anthropic's Fable model to build K3, alleging Moonshot developed an internal platform to rapidly switch between access methods while avoiding detection and accessed banned NVIDIA GB300 Blackwell chips through Thailand-based infrastructure. No supporting logs or evidence have been made public. Treasury Secretary Scott Bessent separately threatened sanctions and possible Entity List blacklisting for Chinese AI companies that extract capabilities from American models. The Bureau of Industry and Security is formally investigating. Moonshot has not publicly addressed the July 22 allegations.

China's Export Control Consultations

On July 21, 2026, the Financial Times reported that China's Ministry of Commerce had been consulting Alibaba, ByteDance, and Zhipu on restricting foreign access to AI model weights and training data. The measures, if adopted, would enter China's technology export catalogue as restricted technologies requiring licenses. No decision has been finalized, and industry feedback from the consulted companies has reportedly pushed back on the controls, citing risks to China's own AI development. The K3 weights release scheduled for July 27 would be difficult to retract regardless of subsequent policy decisions.