Three stories worth your coffee today. One AI lab put a scary sentence in an SEC filing, a pile of newspapers sued a browser company for running a content chop shop, and AMD posted a quarter that shows who is actually getting paid in this boom.
01 of 03
Anthropic says, in an SEC filing, that self-building AI may be close
Anthropic filed a revised S-1/A on Monday, November 2, and buried in the risk factors is the most interesting sentence any AI company has put in front of regulators this year. The company says its "internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy's definition of 'dramatically accelerated progress attributable to AI.'"
That threshold is what most people call recursive self-improvement, or RSI. It is the point where AI gets good enough at building AI that progress speeds up on its own. Anthropic is careful to say no current model has crossed that line, and that it is still assessing the data.
The part that got my attention is the admission about the past. The filing says the company's growing use of AI to speed up its own research "may cause the capabilities of our models to advance more rapidly than our ability to develop, validate, and implement corresponding safety, alignment, and security measures," and that it has "in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred."
Anthropic also says it does not expect to be able to safely let Claude "fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding," and that it is considering updates to its Responsible Scaling Policy to keep humans in control.
Here is where I land on it. Plenty of folks will call this hype dressed up as humility, and some skeptics already have. But this is an S-1. Lawyers read every word, and you can get sued for what you put in there. When a company tells future shareholders "we might not notice the speedup until after it happens," that is not a marketing line. That is a company covering itself for a risk it thinks is real. I would take it more seriously than any keynote.
Source: Anthropic S-1/A on SEC EDGAR, via TechCrunch
02 of 03
The New York Times calls Brave an AI "fence" for stolen content
The New York Times sued Brave on Monday in federal court in Manhattan. The same day, the Daily News, the Chicago Tribune, and the Orlando Sentinel filed a similar suit in the same court.
The claim is that Brave, the 11-year-old privacy browser company, scraped their copyrighted journalism and sold it to AI companies. The suit names five customers: Anthropic, Perplexity, Mistral, xAI, and Cohere. According to the complaint, Brave built an index of more than 30 billion web pages with a crawler it never identified, so publishers had no way to block it.
The Times did not pull punches on the wording. The suit says Brave "acts both like a burglar that misappropriates the content of publishers, and an AI 'fence' that sells stolen content to third parties." It also goes after a Brave browser feature that gets around paywalls, and Brave's habit of blocking publishers' ads while selling its own. The papers want an injunction and damages. Brave had not commented as of the first reports.
The Times has already sued OpenAI and Perplexity over similar claims.
What I want to know is whether this is the smarter target. Suing the AI labs means fighting over fair use and model training, which is a slow, messy argument. Suing the middleman that supplies search results to those labs is a cleaner story: you took our pages, you resold them, pay up. If that theory holds, every company selling a search API to AI shops is going to be reading this complaint very closely.
Source: TechCrunch
03 of 03
AMD's data center business nearly doubled, and the bill for its big deals showed up
AMD reported third quarter results Tuesday. Revenue was a record $13.4 billion, up 45% from a year ago. The data center segment did $8.5 billion, up 97%, driven by EPYC server chips and the launch of the Instinct MI450 GPUs. Non-GAAP earnings were $2.09 a share, up 74%. For the fourth quarter, AMD guided to about $17 billion, give or take 3%, which would be roughly 66% growth year over year.
The wrinkle is the warrants. AMD gave OpenAI and Meta the option to buy up to roughly 10% of the company each, tied to chip purchases. Meta's comes with a deal to buy up to $60 billion of AI chips over five years. AMD's release shows a $1.1 billion hit, or 66 cents a share, from remeasuring those warrants this quarter. Per Reuters, OpenAI's warrant also has share-price thresholds and Meta's does not, and both need shipment and other commercial conditions met before they vest. The stock still rose about 8% after hours.
Lisa Su said AMD expects "growth to accelerate further in the fourth quarter as we scale Helios and 'Venice' production."
My read is simple. AMD basically paid its two biggest customers in stock to pick it over Nvidia, and now the cost is showing up every quarter as the share price climbs. That is a fine trade if the chips keep shipping. But it tells you who has the leverage in this market. It is not the chipmaker. It is whoever has the biggest data center budget.
Sources: AMD press release, Reuters via WTAQ