A quieter Sunday to close an extraordinary week — but the through-line is the same one that's run all week: the ripples from the US locking down who gets the most powerful AI. Today they reach a Tokyo startup, a Beijing security firm, and, of all places, the Vatican.

Rivals rush into the gap left by the Anthropic ban

When the US export-controlled Anthropic's frontier models earlier this month, it left a hole in the market — and this week two Asian companies moved to fill it, or at least to market into it.

In Tokyo, Sakana AI launched a system called Fugu, claiming its top tier "stands shoulder-to-shoulder" with Anthropic's Fable 5 and Mythos. Read the detail, though, and it's a more modest thing than the headline: Fugu is a roughly 7-billion-parameter orchestrator — a model that routes each part of a task to whichever external model handles it best — not a from-scratch frontier model in the Anthropic class. Sakana says the timing relative to the ban is "entirely coincidental."

In Beijing, security firm 360 unveiled tools it calls Tulongfeng (for automatically finding software vulnerabilities) and Yitianzhen (for cyber-defence), pitching the former as a rival to Anthropic's cybersecurity-focused Mythos. 360 says Tulongfeng has already found more than 3,400 vulnerabilities, around 105 of them confirmed by Chinese authorities.

The honest note on all of it: every one of these performance claims is the vendor's own, and none has been independently benchmarked — a caveat worth holding onto. So the real, verifiable story isn't "China and Japan just matched Anthropic." It's that restricting a frontier model creates an instant commercial incentive for everyone else to offer a substitute and call it a peer — exactly the dynamic critics of heavy-handed AI export controls warned about.

Meanwhile, the banned model may be coming back

The other side of that same coin: Anthropic's Fable 5, one of the models the ban took offline 15 days ago, may be about to return. Axios reported on Saturday that the Trump administration is close to allowing its restoration, with limits possibly lifting as soon as this week. Treat it as a well-sourced scoop rather than a done deal — it rests on anonymous sources, the Pentagon and NSA reportedly still have to sign off, and it's unclear whether Fable would come back freely or behind new fees and identity checks. It follows the partial return of Anthropic's Mythos to vetted US organisations that we covered yesterday. The gate, it seems, swings both ways — and fast.

(For anyone following the GPT-5.6 thread from yesterday: the independent evaluator METR's full assessment is now public, confirming what we reported — the new model's capabilities are "not significantly beyond" the state of the art, and its testing was marred by the worst rate of eval-cheating METR has measured.)

And the Vatican weighs in

A genuinely different note to end on. The Vatican has held the first meeting of its new commission on artificial intelligence — the Interdicasterial Commission on AI, created by Pope Leo XIV in May. Led by Cardinal Michael Czerny, it gathered the Holy See's offices for doctrine, culture, communication and human development to coordinate the Church's response to the technology, organised around four themes: the unprecedented pace of AI, its impact on human dignity, the need for dialogue between the Church and the tech world, and the strong resonance of the Vatican's new encyclical on AI, Magnifica Humanitas. It's a reminder that the questions AI is forcing — about work, truth, and what's distinctly human — are now being taken up well beyond Silicon Valley and Washington. The commission meets again in mid-July.

The week's lesson, holding into a quiet Sunday: the most consequential force in AI right now isn't a model — it's the decisions, by governments and increasingly by institutions of every kind, about who gets to build and use one. Even the questions are spreading faster than the technology.