Breaking

Five Eyes Warns China, Japan Closing on US Frontier AI With Cheaper Models

June 24, 2026 at 07:30 EDT

Foundation Models
Policy & Regulation
Open Source

The Five Eyes intelligence alliance (US, UK, Canada, Australia, New Zealand) warned in June 2026 that Chinese and Japanese AI firms are rapidly closing in on American frontier AI using far cheaper models, framing the timeline as "not years, it is months." According to Axios, China's open-source strategy and Japan's orchestration approach are narrowing the capability gap the US has long held faster than intelligence agencies anticipated. In a joint statement, Five Eyes cast the shift as an accelerant of cyber threats and called for a "whole-of-society response."

June 22, 2026 · Five Eyes Cyber Agencies

"The Timeline Is Not Years, It Is Months"

A rare joint warning from the U.S., U.K., Canada, Australia and New Zealand says frontier AI will reshape cyber offense and defense — as China and Japan close on U.S. frontier models far faster, and far cheaper, than expected.

intelligence agency heads signed the joint statement

~7 mo

average gap by which Chinese models trail U.S. — and shrinking

~20×

cheaper: ~$0.50 vs ~$10 for the same coding task

Output cost · per million tokens

Same-tier results, a fraction of the price

$30

Sakana AI
Fugu Ultra

~$4.40

z.AI
GLM-5.2

Open-weight GLM-5.2 prices output at roughly one-seventh of Fugu's tier while competing on benchmarks.

Closing the gap on benchmarks vs GPT-5.5

SWE-bench ProGPT-5.5: 58.6

GLM-5.2: 62.1 — surpasses GPT-5.5

FrontierSWEGPT-5.5: 72.6%

GLM-5.2: 74.4% — edges ahead on long-horizon coding

Japan · Sakana AI "Fugu Ultra"

An orchestrator that routes tasks across public models via one API — claims "frontier capability without export-control risk," standing shoulder-to-shoulder with Fable 5 and Mythos Preview.

China · z.AI "GLM-5.2"

Open-weight (MIT), 1M-token context, strong long-horizon agentic coding — competing with GPT-5.5 at a fraction of the cost.

What the agencies urge — defenders, act now

Reduce attack surface Accelerate patching Address legacy systems Strengthen access controls Prepare for incidents Use AI for defense

The window between vulnerability discovery and exploitation is shrinking — distillation and open-weight proliferation raise the stakes, while export controls may push users toward non-U.S. models.

Continue reading

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