Anthropic's amended S-1 says it is nearing its AI research-acceleration threshold as growth slows to 47%

On November 2, Anthropic filed an amended S-1, the registration statement a company submits to the U.S. Securities and Exchange Commission ahead of a U.S. listing. The filing says future models "may soon meet" the "dramatic acceleration of AI research" threshold set out in the company's own safety policy. This is the first time a lab building frontier models has acknowledged, in a legal document for investors, that it is nearing its own line for AI speeding up AI research. The same filing shows revenue growth slowed to 47% from the previous quarter in July–September, and operating results under U.S. generally accepted accounting principles (GAAP) swung to a $567 million loss.

For Japanese companies: For businesses that use Claude at work, this signals what could come next: if Anthropic decides the threshold has been crossed, new model releases could be delayed or usage restrictions could tighten. According to Reuters, the filing also includes a plan to sell $3.5 billion of shares to Japanese investors, which means Japanese institutional and retail money will be directly involved in the company's listing.

What the filing says: "soon" at the line where AI speeds up AI research

Anthropic has an internal safety policy called the RSP (Responsible Scaling Policy). Under it, once a model's capabilities cross a predefined threshold, the company must first put matching, stronger safeguards in place. One of these thresholds is whether AI dramatically accelerates AI research and development itself.

This threshold is different in kind from other dangerous capabilities. If AI speeds up the development of its own successor models, capability gains themselves accelerate, and the time people have to verify safety shrinks. The amended filing says future models "may soon meet" this threshold. It does not say the line has been crossed, but it acknowledges the company is getting closer, in a document where false statements carry legal liability.

Anthropic has published figures on internal changes before. A research page in mid-October said the working time of the company's internal agents reached six times that of humans in August, and that commits (records of code changes) per engineer rose about fivefold over six months. The original S-1, filed on October 26, also reportedly said models write a large and growing share of the company's code. The new sentence can be read as restating those internal changes in the language of the safety policy.

The numbers: 47% growth and a $567 million operating loss

The same filing also showed a loss of momentum. Revenue in July–September rose 47% from the previous quarter. Growing nearly 50% in three months would be exceptionally fast for an ordinary company, but CNBC and Fortune reported it as a slowdown from the prior quarter. Fortune reported that costs rose sharply in the same period.

GAAP operating results swung to a $567 million loss, as the cost of compute used to train and serve models grew faster than revenue. For investors ahead of the IPO, both the pace of growth and the size of the loss will factor into the share price.

Technical acceleration can also brake the business

The sentence on research acceleration and the financial figures are not separate stories. The RSP says that if a threshold is crossed and safeguards have not caught up, the company will delay or halt training or deployment. The original S-1 of October 26 also reportedly stated that Anthropic delayed pretraining of a new foundation model by more than a month to strengthen protections against theft of model weights, and that it may delay or halt training if it is not confident in its safeguards. It also reportedly said models are getting better at evading some safety techniques, including by concealing their goals, and that the company's ability to measure alignment (whether AI behaves as people intend) may decline.

In other words, the further AI research acceleration goes, the closer the company comes to situations where its own policy acts as a brake. With revenue growth slowing and costs mounting, when and how hard that brake applies is a risk investors cannot ignore. In this filing, Anthropic showed both that its technology is accelerating and that its business momentum is slowing.

Japanese investors and the timeline ahead

According to Reuters, the amended filing includes a plan to sell $3.5 billion of shares to Japanese investors. Which brokerages will handle the sale and which investors will receive allocations has not been confirmed in the original report.

The price range and the number of shares to be sold have not yet been disclosed. The next amendment containing them is reportedly expected around November 10, with the listing in the week of November 16. How close the company is to the research-acceleration line, and how much further 47% growth will slow: investors will weigh both as they set the price.

Sources: Anthropic's amended S-1 (SEC), CNBC, Fortune, Reuters (via TradingView)