Gather — 29 July 2026 · Zeitgeist — a spike by Chris Gathercole

The EU Digital Omnibus Agreement and AI Act Article 53: Reshaping Copyright Licensing for General-Purpose AI Training (IP and Legal Filings, 2026) — Following the May 7 trilogue agreement on the Digital Omnibus package, this analysis confirms Article 53’s copyright-compliance and training-data-summary obligations remain anchored to the August 2, 2026 deadline this journal has tracked since May — even as the same Omnibus deal pushes back separate high-risk AI Act obligations. GPAI providers must still implement copyright-compliance policies and publish training-data summaries via the SEND platform on the original schedule.
Digital Omnibus on AI Provisional Agreement Reached at the May Trilogue (Bird & Bird, 2026) — Primary-adjacent confirmation of the May 7 political agreement between Parliament, Council, and Commission: the Omnibus selectively defers high-risk AI system obligations while leaving GPAI transparency/copyright provisions untouched — a deliberate triage decision, not an oversight, three days ahead of the deadline’s activation.

South Korea issues guidance on copyright and AI training data (Asia IP, 2026) — South Korea’s Ministry of Culture, Sports and Tourism and Korea Copyright Commission guidance explicitly excludes AI-training-phase disputes from its Copyright Dispute Prevention Guidelines, leaving training-data infringement to be addressed separately — even as South Korea’s Copyright Act still lacks a text-and-data-mining exception. This closes part of the gap flagged repeatedly since April: Korea is now the second Asia-Pacific jurisdiction (after Japan, 07-27 gather) with a documented policy position, and it is notably less settled than Japan’s Article 30-4 exception.
China: First AI output copyright infringement case (Linklaters, 2026) — The Guangzhou Internet Court’s “Ultraman case” (Shanghai Xinchuanghua v. Guangzhou Nianguang) found a generative-AI image service indirectly liable for outputs closely resembling copyrighted characters, on a “knew or should have known” standard — China’s first ruling on AI-output infringement, though (unlike the US/EU cases tracked here) it addresses output similarity rather than training-data sourcing. First substantive China data point in this journal; training-data-specific litigation there remains unconfirmed.

Scholarly Publisher AI Licensing Deals: Inside the Numbers (CASRAI, 2026) — Academic publishers are now reporting AI licensing revenue in investor-facing fiscal results rather than one-off deal announcements: Wiley disclosed $49M in FY2026 AI licensing revenue (lifetime total over $110M); Taylor & Francis’s Microsoft deal is running at ~$10M in year one with total AI-related revenue expected to exceed $75M portfolio-wide; Springer Nature’s 2024 Google deal ($23M one-time) functions as the benchmark valuation other negotiations reference. Large publishers with well-cleared back catalogues are capturing recurring revenue; the long tail of small/society publishers reports little or nothing — extending the “real-time access, not one-time acquisition” and sector-diversification patterns already tracked here (media, 06-26; biopharma, 07-23) into academic publishing specifically.

AI Training and Copyright in Europe: A Potential Shift Beyond Territoriality (Potomac Law Group, published 2026-03-19, surfaced this cycle) — Argues the EU’s text-and-data-mining exception is being applied extraterritorially: once a foundation model is placed on the EU market, EU copyright law’s treatment of its training data can apply regardless of where the training itself occurred, following the November 2025 Munich GEMA v. OpenAI logic that the TDM exception did not cover that training. If this reading holds, “train outside the EU” stops being a jurisdictional escape hatch for any lab wanting EU market access — a novel legal theory not previously tracked in this journal despite being in circulation since March.

AI Impact on Society The Digital Omnibus’s selective deferral — postponing high-risk AI Act obligations while explicitly preserving the August 2 GPAI copyright-transparency deadline — is a concrete data point on how EU regulators are triaging AI Act enforcement priorities, distinct from the training-data substance itself.
Open vs Closed AI Ecosystems EU copyright’s alleged extraterritorial reach (Potomac Law) would remove training location as a variable for open-weight developers seeking EU distribution — training outside the EU would no longer avoid EU copyright exposure once a model reaches the EU market.

Gap: The repeated Asia-Pacific gap (flagged 04-05, 04-25, partially closed 07-27 with Japan) narrows further with South Korea and China data points this cycle — though China’s is an output-similarity ruling, not a training-data case, and Korea’s guidance explicitly punts on training-phase disputes. Neither closes the gap as fully as Japan’s Article 30-4 framework does.
Emerging theme: Academic/scholarly publishing is the second sector (after biopharma, 07-23) to shift from speculative deal-announcement coverage to disclosed, investor-facing licensing revenue figures — Wiley and Taylor & Francis are now reporting AI licensing income the way they’d report any other revenue line.
Method note: The Potomac Law extraterritoriality piece is four months old but surfaces a legal theory absent from this journal’s prior coverage — a reminder that routine keyword sweeps can miss substantive analysis that doesn’t use “lawsuit” or “ruling” framing; periodic broader sweeps for novel legal theories (not just case-status updates) would catch this faster.
Quality signal: CASRAI’s use of disclosed fiscal-year figures (rather than trade-press deal-value estimates) is a materially higher-confidence data source than the deal-announcement coverage this journal has relied on for the licensing-market thread; prioritize investor-disclosure-sourced figures where available.