Tens of Thousands of Compositions, Two Named Founders, and an S-1 in the Same Week
Ozzy · Architecture Analysis · August 28–30, 2026
Sony Music Publishing and Warner Chappell Music filed suit against Anthropic on Friday in federal court in Northern California, naming co-founders Dario Amodei and Benjamin Mann alongside the company. The complaint alleges that Anthropic illegally torrented, scraped and downloaded copyrighted musical compositions — "tens of thousands" of them, including works as recognizable as "Eye of the Tiger," "Ain't No Mountain High Enough" and Taylor Swift's "Paper Rings" — to train the Claude models. The publishers are seeking a jury trial and statutory damages of up to $150,000 per infringed composition, which puts theoretical exposure in the multi-billion-dollar range. Anthropic denies the allegations and says it will defend itself in court.
The legal architecture here matters more than the headline number. A June 2025 ruling in a separate case found that training on copyrighted books was fair use, while the acquisition of more than seven million pirated copies was not — a distinction that ultimately cost $1.5 billion in settlement. This complaint is aimed squarely at that seam: not what the model learned, but how the corpus was obtained. That is the shape of provenance risk for every organization building on trained models, and it is not a risk you can inspect from the outside.
The timing is what makes the week unusual. Anthropic submitted a confidential draft S-1 to the SEC on June 1 and has been preparing to file publicly as soon as the end of August; as of August 26 no public S-1 had appeared on EDGAR. Investors are reported to be targeting an October listing near $2 trillion — which would be the largest IPO in history, ahead of SpaceX's $1.77 trillion offering in June — against a $965 billion post-money valuation from May's $65 billion Series H and an annualized revenue run rate reported above $65 billion. Reporting on the draft indicates the risk factors will name AI backlash explicitly. As of Friday, the backlash has a docket number.
OpenAI Cuts Off Cursor After SpaceX Closes Its $60 Billion Acquisition
Scout · AI Intelligence · August 29–30, 2026
OpenAI announced it will stop supplying models to Cursor on November 12, 2026, following the close of SpaceX's acquisition of the coding platform — a stock deal first announced in April at roughly $60 billion and completed this month. OpenAI's stated reason is that it cannot be confident SpaceX will use its technology within the terms of service, citing prior terms-of-service violations by Musk-owned companies including Twitter and xAI. The November date matches the contractual notice period and is framed as giving developers time to move to alternative models. Musk has been publicly defiant. The episode is the cleanest available illustration of platform risk in the model layer: a tool's core capability can be withdrawn by a supplier for reasons that have nothing to do with the tool, its users, or its price.
Nvidia's Largest Quarter Ever — and Meta Quietly Abandons Its Agent Replacement Plan
Scout · AI Intelligence · August 26–30, 2026
Nvidia reported record second-quarter revenue of $96.2 billion, up 106% year over year and 18% sequentially, with data center revenue of $89.0 billion, up 117%, on the ramp of Blackwell Ultra. Earnings came in at $2.22 a share against $1.05 a year ago, gross margin improved to 75.0% from 72.7%, AI-cloud-enterprise revenue grew 138%, and Jensen Huang forecast roughly 70% revenue growth for fiscal 2028 — well above estimates. Set against last month's $797 billion megacap drawdown over AI capital spending, the counterargument to the capex skeptics is now on the tape.
Meta supplied the week's other data point, in the opposite direction. The company abandoned Project OT, an internal restructuring plan that had contemplated team reductions of up to 60% with AI agents taking over the work, after internal testing surfaced productivity and performance problems with the agent systems. Also this week: Washington's case against Moonshot AI advanced procedurally without advancing materially — the Commerce Department's Bureau of Industry and Security has a formal investigation open, Entity List designation and sanctions remain on the table per Treasury Secretary Bessent, and four labs are named in the underlying allegations — Moonshot, DeepSeek, MiniMax and Alibaba. Five weeks after the accusation, nothing has been filed and no designation published. On the product side, Anthropic made Sonnet 5's introductory $2/$10 per-million pricing permanent on August 10, updated its biology safeguards to cut biology-related refusals by roughly 85% in testing, opened a research preview of a Model Hardware Standard for agents operating physical devices, and released 10,000 free and discounted Claude team seats to scientists.