Monday, 7 September 2026

Some authors owed a share of Anthropic's $1.5 billion copyright settlement got emails this week saying someone else was claiming their money [1]. Anthropic settled the class action last year, after a judge ruled that training AI on copyrighted books is fair use but pirating them is not; the deal won final approval in July [1]. Under its terms, the authors of nearly 500,000 titles are paid $3,000 for each pirated work — split 50-50 with the publisher if the book is still in print, but paid in full to the author if it was self-published or the rights have reverted [1].

Writers report two recurring errors: publishers claiming books whose rights reverted years ago, and publishers claiming the full payment when they are owed half [1]. Mystery author April Henry said HarperCollins claimed a book that reverted "at least 17 years ago" [1]. Victoria Strauss of Writers Beware said the volume and sameness of complaints suggest something "widespread and systemic," though she is "reluctant to attribute to malice" what bad recordkeeping explains [1]. Authors Guild CEO Mary Rasenberger told The New York Times she does not see it as "a grab by the publishers" [1]. Literary agencies are also filing claims, which Strauss called surprising since "agents are not rightsholders" [1]. According to [2], first payments are due between November 1 and 15.

If you write or publish, this is a preview of how a won case actually pays out: the money exists, but who is owed which slice is contested, and the paperwork decides it.

Buyers are calling it model fatigue: the pace of releases has outrun anyone's ability to evaluate them [1][2]. In one week Anthropic updated Claude Fable 5.1 and Mythos 5.1, Meta announced Muse Spark 1.3, Google unveiled Gemini 3.8 Flash, and OpenAI shipped GPT-6 Astra, while Abu Dhabi's MBZUAI released its open-source K2 Horizon family [1]. "I feel like model fatigue is a real thing," Zhen Lu, CEO of the startup Runpod, told CNBC [1].

The numbers behind the feeling are stark. According to [2], the median gap between major model releases has compressed from 37.5 days in 2023 to 11 days so far in 2026; for OpenAI alone it fell from 170.5 days to 49. In July 2026, more than 1,000 employees across major labs signed a petition calling for a more measured pace [2]. Anthropic and OpenAI, both valued near $1 trillion by private investors and both heading toward public markets, are pushing hardest [1]. One analyst told CNBC that most of this week's launches were "point releases" — upgrades to existing models — and that the last models to truly move the needle were Anthropic's Fable 5 in June and Moonshot's Kimi K3 in July [1].

For a buyer, the read is not to chase each release. When four labs ship in a week and benchmarks converge, the durable skill is a routing rule — which task goes to which model, and when it is worth switching — not loyalty to whatever launched last.

An Ars Technica investigation into the $3.2 billion Lake Mariner campus in Somerset, New York — a former coal mine on Lake Ontario — shows how diffuse responsibility becomes when an AI data center is built [1]. TeraWulf owns and operates the site on land leased from a company owned by its own CEO; the UK's Fluidstack will run it; Google holds warrants for a future 14 percent equity stake and guarantees Fluidstack's lease payments; and Anthropic is among the AI firms whose compute demand it exists to serve [1].

The gap between promise and delivery is wide. TeraWulf's own planning presentation put full-buildout employment at 35 to 40 people for a facility drawing up to 500 megawatts — against the 165 permanent jobs and $85 million in capital investment promised in 2019 when the landlord entity sought a power discount [1]. After a fire at the site, firefighters reportedly found three dead hydrants, and the safety documents they are legally entitled to see had burned up in the blaze [1]. In February 2026, Anthropic pledged to cover electricity price increases from its data centers, including 100 percent of grid infrastructure costs, and says Lake Mariner is covered — but the pledge addresses pricing, not whether clean-energy or noise commitments are being kept [1]. According to [2], New York Governor Kathy Hochul signed the country's first statewide moratorium on new hyperscale data centers in July, pausing permits for up to a year.

The compute behind your AI tools is built by layered companies, and when its costs land on a community, no single one is clearly on the hook.

OpenAI, the World Association of News Publishers (WAN-IFRA) and the Association of Independent Regional Press Publishers of Ukraine announced on 7 September a program to help ten Ukrainian regional newsrooms adopt AI [1]. Participating publishers get OpenAI API credits and hands-on help building workflows, split across an editorial track and a commercial-and-operational one [1].

The program runs in two stages. According to [1], the Newsroom AI Masterclass Series began on August 5, bringing in international practitioners on workflows, audience, product and revenue; the Newsroom AI Catalyst launches September 17 and gives deeper support to ten publishers chosen by the Ukrainian association to pick high-impact uses, build roadmaps and pilot custom tools. OpenAI has worked with WAN-IFRA since 2024, and WAN-IFRA has backed Ukrainian media since Russia's full-scale invasion in 2022 [1][2]. Oksana Brovko, who leads the Ukrainian publishers' association, framed the goal as giving journalists back time to do journalism [1].

The wider lesson is the one every small team faces when it adopts AI. Credits and training start a pilot; connectivity, staff time and someone to maintain the tools after the program ends are what decide whether it survives. Ten pilots will show quickly which uses stick — a small, honest test worth watching if you run a lean content or service business.