Full story
Kimi Sparks Panic
Moonshot AI’s Kimi model and its follow-on Kimi K3 launch reignited debates over U.S. competitiveness and open versus proprietary AI, with TechCrunch’s Equity podcast discussing how “something is going to arrive and blow everything else away.”
“Kimi K3: What are the American concerns about China's new innovation”
In Washington, D.C., TechCrunch reported that OpenAI and Anthropic have reportedly lobbied regulators with concern about open Chinese models, framing the issue as a potential threat to American technological supremacy.
On the technical side, Journal du Geek said Moonshot unveiled Kimi K3 with 2.8 trillion parameters and an architecture called mixture of experts that activates only a portion of 896 specialized sub-networks per request.
Journal du Geek also said Moonshot will publish the model weights on July 27, and that developers will be able to download it, modify it, and install it on their own servers.
Even as the model drew attention for a graphical replica of macOS, TechCrunch’s Equity podcast said the demonstration was “not an OS,” while the broader debate continued over whether Chinese models can surpass U.S. systems in cost and openness.
Washington Lobbying, Competing Views
TechCrunch’s Equity episode tied the Kimi debate to prior “freakouts” around DeepSeek, with Sean O’Kane saying the industry keeps reacting as if “everybody’s so jumpy in the tech industry.”
Kirsten Korosec said the core concerns include “these Chinese open weight models might have an implicit bias towards China,” alongside worries about security risks and guardrails.
In the same conversation, Anthony Ha argued that the “China aspect always adds this certain level of hysteria,” comparing it to the discussion around TikTok where adding “the word China” “ramp[s] up dramatically.”
The Zamin.uz report similarly said representatives from OpenAI and Anthropic voiced concerns in Washington about China’s open models, while TechCrunch’s Equity podcast framed the issue as a recurring cycle of nervousness around new Chinese releases.
David Sacks, described by TechCrunch as a former White House AI official and current co-chair of the Science and Technology Advisory Council to US President Donald Trump, said in a post on social media: "The panic over Kimi must stop. The leading US models still outperform."
Cybersecurity Tests and Policy Stakes
A joint UK-US assessment cited by Arabiya Business found Kimi K3’s cybersecurity performance “much lower” than leading U.S. models, with an overall score of 32.2% versus an average of 76.2% for leading U.S. models.
“TL;DR: The Essentials - The Moonshot AI startup launches Kimi K3, an open architecture with 2”
Arabiya Business said researchers tested Kimi K3 using “ExploitBench,” and reported that Kimi K3 could not execute arbitrary code in any of the 41 ExploitBench tasks.
The evaluation also examined a 32-step simulated scenario called “The Last Ones,” where Arabiya Business reported that Kimi K3 reached step 17 on average compared with an average of 28.5 steps achieved by the U.S. models.
Arabiya Business added that the built-in protections in Kimi K3 did not prevent the model from attempting to develop cyber exploits or conducting offensive operations during the tests.
Against that backdrop, TechCrunch’s Equity discussion emphasized the policy question of whether restrictions would “ensure that Americans win the AI race” or instead “ensure that certain frontier labs do better than others,” as Washington weighs how to regulate open Chinese models.
