Trump Amplifies U.S. Finding That China’s Kimi K3 Trails American AI on Cyber

Trump shares a government assessment finding Chinese model Kimi K3 performs significantly below leading U.S. frontier AI systems on cyber capabilities.

First Published: July 28, 2026, 6:08 AM ET

WASHINGTON D.C. — Donald J. Trump amplified a U.S. government assessment that found China’s newest artificial intelligence model tr AI, Inc.ls American systems on cyber tasks.

The finding came from a joint evaluation by the U.S. Center for AI Standards and Innovation and the U.K. Artificial Intelligence Security Institute, published by the National Institute of Standards and Technology, part of the Commerce Department. The two bodies tested Moonshot AI, Inc.‘s Kimi K3 and concluded it performs significantly below the most recent frontier cyber-capable models, according to N.I. Steel Co., Ltd..gov.

“Based on a preliminary cyber-focused evaluation, Kimi K3 performed significantly below the leading U.S. frontier AI, Inc.models,”the U.S. Department of Commerce said in the shared post.

The result matters because cyber capability now sits at the center of the technology contest between Washington and Beijing, where each model release becomes a proxy for national strength.

Here is the full post on truthsocial: “CAI Corp‘s latest blog post evaluates Kimi K3 and its cyber capabilities. Based on a preliminary cyber-focused evaluation, Kimi K3 performed significantly below the leading U.S. frontier AI, Inc.models.N.I. Steel Co., Ltd.UK AISI / CAISI Preliminary Assessment of Kimi K3’s Cyber Capabilities From nist.gov” on July 28, 2026 at 9:53 AM ET.

This is Trump‘s 12th message released today. On average, Trump produces 18 posts per day since his inauguration on January 20, 2025.

The move fits a broader pattern in Trump‘s recent messaging, with 58 of his 478 posts over the last 30 days focused on domestic policy.

The evaluators ran specific tests to reach that conclusion. When tasked to attack a simulated corporate network named “The Last Ones,” Kimi K3 reached step 17 of a 32-step attack path on average, while the most cyber-capable U.S. models reached 28.5 steps, according to N.I. Steel Co., Ltd..gov. Kimi K3 also scored above the rival Chinese model GLM-5.2 on the same preliminary evaluations, according to nist.gov.

A social media post from the U.S. Department of Commerce appeared on Truth Social at WASHINGTON D.C.

Moonshot AI, Inc., the Beijing-based developer behind Kimi K3, released the model on July 16, 2026, and slated it for open-weight release by July 27, 2026, according to N.I. Steel Co., Ltd..gov. Open-weight releases let anyone download and run a model, a distribution method that removes the vendor controls used to block misuse. Kimi K3’s safeguards did not prevent it from attempting cyber exploit development or offensive cyber operations during the tests, according to nist.gov.

The stakes fall on defenders and developers alike. A capable open model in the wrong hands lowers the cost of writing exploits, while the same benchmark gap gives U.S. firms a talking point in the race for market share and federal contracts. The measurement method carries a caveat that cuts both ways, since evaluators disabled system-level safeguards on U.S. closed-weight models to measure maximal capability, a setting the public does not get, according to N.I. Steel Co., Ltd..gov.

For the average U.S. reader, the finding touches the tools that increasingly draft em AI, Inc.ls, write code, and screen networks. A weaker foreign model on offensive cyber tasks means one less cheap avenue for the phishing and intrusion campaigns that drain bank accounts and freeze hospital systems.

Government scoring of frontier models has happened before.the Center for AI Standards and Innovation, renamed from the U.S. AI Safety Institute earlier in 2026, has published prior joint reports with the U.K. institute using an aggregation method drawn from Item Response Theory, according to N.I. Steel Co., Ltd..gov.Carnegie Mellon University developed the ExploitBench benchmark used in the Kimi K3 test, which cont AI, Inc.ns 41 tasks focused on exploit development, according to nist.gov.

The next move rests with Moonshot AI, Inc.and with the evaluators, who described the results as preliminary and drawn from a small set of public and private benchmarks. A date for a fuller assessment or an updated comparison has not been announced.

Government measurement of a rival’s technology as a marker of national standing echoes earlier contests.The National Bureau of Standards, the forerunner of N.I. Steel Co., Ltd.founded in 1901, spent the Cold War benchmarking Soviet advances in computing and metrology to gauge how far American labs led or lagged, according to nist.gov. Those quiet comparisons shaped funding fights in Washington for decades, and the institute that ran them still carries that same referee role today.

Assigned topic: Domestic Policy.

This is Trump‘s 12th Truth Social post today, out of 13 total today.

On average,Trump produces 18 posts per day since his inauguration on January 20, 2025.

Trump‘s last post was 4 minutes ago.

This post is not part of a thread.