Five Eyes Warn Frontier AI Risks Are Imminent; Chinese and Japanese Models Closing Gap Faster Than Expected
Miles Bennett
The Five Eyes intelligence alliance issued a rare joint warning this week: frontier AI capable of severely damaging governments and enterprises is months away, not years. The trigger is that Chinese and Japanese low-cost models are catching up far faster than intelligence agencies anticipated, narrowing the U.S. lead in real time.
What exactly are the Five Eyes warning about?
Intelligence agencies of the U.S., U.K., Canada, Australia, and New Zealand issued a joint alert: frontier AI that can inflict serious harm on governments and businesses is "months, not years" away.
This means → the intelligence community has reclassified frontier AI from a long-term concern to a near-term threat — an outright escalation in urgency.
The alliance called for a "whole-of-society response" and singled out corporate boards: deploying controls is not enough — those controls must hold under real pressure.
Why did Chinese and Japanese models catch intelligence agencies off guard?
Anthropic's Mythos is currently viewed as the world's most potent AI cyber-threat. OpenAI is very close. But China and Japan are closing the gap at significantly lower cost.
Zhipu's GLM-5.2 ties GPT-5.5 on the Artificial Analysis leaderboard, yet runs at one-fifth the cost. In plain terms = same job, one-fifth the bill.
Japan's Sakana AI launched Fugu Ultra — a model router that intelligently switches among publicly available models. The company claims it delivers "frontier capability with no export-control risk" at near-Mythos performance. This signals a new playbook: instead of building the single strongest model, combine existing ones to sidestep controls.
The "model distillation" controversy — the dark side of low-cost catch-up?
In February, Anthropic accused DeepSeek, Minimax, and Moonshot of using thousands of accounts and millions of interactions to illegally distill knowledge from Claude.
In plain terms = model distillation — training a weaker model on a stronger model's outputs — lets rivals replicate billions of dollars in R&D at a fraction of the cost.
This means → even if U.S. labs stay technically ahead, knowledge may leak faster than models can iterate, eroding the lead through shortcuts rather than breakthroughs.
Is the U.S. caught in its own dilemma?
Two weeks ago the Commerce Department placed export controls on Anthropic's Fable 5 and Mythos 5. Anthropic immediately shut off all external access.
This reflects a core contradiction: tighter controls → more closed models → stronger incentive for rivals to build their own or distill → controls undercut themselves.
Former Facebook chief security officer Alex Stamos put it bluntly: "It is both arrogant and foolish to assume we have the best stuff just because we're Americans."
What is Europe doing?
Milan-based AI company Domyn last week unveiled Project Europa — a frontier open-source AI model supporting all 24 official EU languages.
Domyn also partnered with Nvidia to build Europe's largest AI supercomputer, Colosseum.
This means → the frontier-AI race is no longer a two-player game between the U.S. and China. Europe is carving out a third path — open-source plus multilingual — to avoid being sidelined in the U.S.–China controls standoff.
It is entirely possible that China has very, very powerful models. It is both arrogant and foolish to assume we have the best stuff just because we're Americans.
Alex Stamos
Former Chief Security Officer, Facebook
(Interview with Axios)
Content is for reference only, not financial advice.