A threshold the company set for itself
Anthropic told the US Securities and Exchange Commission on 2 November that it may be about to cross a line it drew in its own safety policy. In Amendment No. 1 to its Form S-1, the company writes that “our 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.’” It adds that it has not designated any current model as exceeding that threshold and “continue[s] to assess the underlying data.”
The filing is unusually candid about the company’s ability to see the acceleration coming. Most of its evidence on the pace of AI-driven progress, it says, comes from lagging indicators, and as a result “we have 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.”
What the first filing did not say
Anthropic filed its original S-1 on 26 October. That version described using each Claude model to help build the next, called the dynamic “closing the loop,” and noted that the risks are covered by its Responsible Scaling Policy as the company approaches “what has been termed ‘recursive self-improvement.’” The passage stopped there.
The amendment extends it. Anthropic now says it does “not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible,” and that it is “considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.” Neither the word “imminently” nor the phrase “keep humans in control” appears anywhere in the October version.
$895.4 billion, where the first draft had a gap
The amendment also fills in figures the original filing left as empty dollar signs. As of and subsequent to 30 September 2026, Anthropic says it has entered purchase commitments and lease agreements — including leases not yet commenced — with an aggregate committed spend of approximately $895.4 billion, substantially all of it compute and infrastructure. About $447.5 billion falls before 31 December 2030, with most of the rest between 2031 and 2045, and $69.8 billion is due within twelve months of 30 September.
Set against that, the contracted revenue in the same document is $13.8 billion of remaining performance obligations, about 75% of which Anthropic expects to recognise within a year. Revenue reached $17.0 billion in the quarter to 30 September, up from $1.4 billion a year earlier, with an operating loss of $567 million and an operating margin of minus 3%.
What to watch
Anthropic says it will publish further reports as its Responsible Scaling Policy requires once it evaluates the implications of meeting the AI research and development capability threshold. Any revision to that policy, and any designation of a model as having crossed the line, would be the first concrete test of language the company has now put in front of prospective shareholders.
