Breaking

White House Ex-AI Adviser David Sacks Says China's Kimi K3 Fixed Security Bugs US Models Wouldn't Touch

July 19, 2026 at 21:21 EDT

Foundation Models
Software Dev & Coding
Policy & Regulation

David Sacks, a former top artificial intelligence official in the Trump administration, has argued that safety restrictions built into leading American AI models are handing a competitive edge to China, citing a case in which Moonshot AI's newly released Kimi K3 fixed critical security flaws that OpenAI's Codex and Anthropic's Fable declined to address.

July 19 · U.S. AI Policy Debate

Do "Cyber Guardrails" Cost America Its Edge?

A former Trump AI official argues U.S. safety limits are backfiring after China's new Kimi K3 fixed 15 critical security bugs that OpenAI's Codex and Anthropic's Fable declined to touch — while shipping bigger, cheaper and open.

critical security bugs Kimi K3 fixed that U.S. models refused

2.8T

parameters — among the largest open-weight models to date

token context window, native multimodal

Price per million tokens

Kimi K3 undercuts both U.S. rivals. Taller stack = higher cost (output tokens).

$15

Kimi K3

$3 in

$30

GPT-5.6 Sol

$5 in

$50

Claude Fable 5

$10 in

Benchmarks: closing the gap

Frontend Code Arena

Kimi K3 #1 1,679

Claude Fable 5 just behind

First Chinese model to top the leaderboard.

Terminal-Bench 2.1

Kimi K3 88.3

GPT-5.6 Sol 88.8

Within half a point of the U.S. frontier.

The Sacks argument

"No reason to limit American models on tasks that Chinese models handle without issue." Guardrails, he warns, are "only making ourselves less competitive" — pushing developers toward open, cheaper alternatives.

The counter-argument

Skeptics question whether benchmarks reflect real work — and reports warn Chinese systems may generate more vulnerable code. The guardrails Sacks criticizes are also meant to prevent genuine security hazards.

What's next

Open weights land around July 27 — letting developers download, fine-tune and self-host Kimi K3 locally, intensifying the debate over whether U.S. cyber guardrails help or hinder competitiveness.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.