Analysis of the AI knowledge economy
The rules, standards, and infrastructure forming around AI training and inference — and what they mean for rights holders and AI builders.
Wikipedia's steward drew the copyright line at retrieval
The Wikimedia Foundation asked a New York court to treat training and retrieval as separate uses, and a license duty at the fetch reaches only rights holders the fetched page can identify.
Read the analysis →French publishers took Google's AI answers to Brussels
Five French publisher organizations joined the Commission's case against AI Overviews and AI Mode, and Google's own documentation ties every refusal to a smaller search result.
Read the analysis →Nordic and Baltic publishers want AI companies to carry the burden of proof
A joint appeal from 14 Nordic and Baltic press organizations asks for consent, a reversed burden of proof, and arbitration, and each demand still turns on naming the work and its owner.
Read the analysis →ChatGPT’s news pipeline has a field for whoever supplied the text
An outside analysis of the engines behind ChatGPT shows supplier identity traveling with the text only where an access deal exists, so the long tail arrives as unnamed excerpts.
Read the analysis →Most publishers that block Claude’s search crawler are almost absent from its answers
Muck Rack data released on October 28 links Claude citations to a robots.txt setting, and that setting is written per domain by whoever runs the server, with no field for the work, its author, or its terms.
Read the analysis →Anthropic’s prospectus prices copyright risk after the use
The S-1 that Anthropic filed on October 26 tells investors that an adverse ruling could mean licensing fees or retraining models from scratch, and each remedy depends on knowing which works went in and who holds the rights.
Read the analysis →Korean books reached AI training by the ton
Woongjin Booxen apologized to Korean publishers on October 21 for supplying 165.69 metric tons of books to Anthropic’s procurement contractor, and the episode shows a purchased copy reaching a model with every record of its rights holder left at the warehouse.
Read the analysis →Europe’s publishers want Meta to ask the rights holder first
Complaints now before the European Commission and the UK competition authority seek to freeze Meta’s crawling terms because a crawl cannot be recalled, and the remedy Europe’s publishers propose rests on a fact no crawler can read today: whose each work is.
Read the analysis →Elvis Presley’s publishers want an accounting of AI training data
A complaint filed on October 23 asks a federal court to make Anthropic account for its training data, and alleges that its cleaning pipeline was chosen for how thoroughly it removed songwriter credits and copyright notices.
Read the analysis →Cloudflare is helping OpenAI find the long tail
Cloudflare’s network helped OpenAI index hundreds of millions of pages that links never surfaced. The signals identify the page and the domain, and the person who wrote the work appears nowhere in the exchange.
Read the analysis →The Anthropic settlement paid only works already on record
Only books registered with the Copyright Office before the use were in the Anthropic class. Now nearly 300 authors are suing Meta one by one, and most specialist knowledge has no record to sue from.
Read the analysis →Canadian publishers are suing Anthropic work by work
A claim filed in Ontario this month invokes a law that prices infringement per work, a remedy only a rights holder with an archive, a catalog of what it owns, and counsel on retainer can use. The long tail has no list of works to sue with.
Read the analysis →The Death of Provenance
For three thousand years an idea could not be separated from the object that carried it. Generative AI is the first medium without that property. Now the answer arrives without a source, and nobody, not even the people who built the system, can say which works produced it.
Read the analysis →Who keeps the record?
Fixed telephony connected places. Mobile communication connected people. IoT connected devices and sensors. Agents are the fourth, and they break the pattern — a device reports, but an AI agent decides, spends and negotiates on someone’s behalf. That moves identity out of the login and into liability. If operators authenticate and wallets identify, the persistent record of agency is still nobody’s job.
Read the analysis →There is no Common Crawl for the physical world
Generative AI was built on data that already existed — the text written, the photographs taken, long before anyone thought to train on any of it. The data physical AI needs mostly does not exist yet, and has to be deliberately produced at real cost. That reframes the whole compensation argument: licensing is not a tax on model training, it is the supply chain of physical AI.
Read the analysis →Two weeks from now, AI licensing stops being optional in Europe
On August 2, 2026, the EU AI Office gains enforcement powers over general-purpose AI: documentation demands, model evaluation, market withdrawal, and fines up to €15M or 3% of global revenue. The obligations have been law since 2025 — now they get teeth. What that means for AI companies and rights holders, including American ones.
Read the analysis →The world’s largest internet market just rewrote the rules for data — and put an end to free AI training
India's DPDP Act is now operational and consent-first. Its AI Governance Guidelines chose a techno-legal path built on provenance standards. And a government copyright working paper rejected free text-and-data mining as a zero-price license — proposing compensated training instead. The direction of travel: AI training as a licensable event.
Read the analysis →Put your knowledge on the record.
Free during the founding period — fingerprinted in your browser; your file never leaves your device.
Create your recordNewsletter
Notes on the AI knowledge economy, written by the founder.
Your address goes into Alltio’s own list and nowhere else.