Anthropic also noted many of its largest clients don’t have long-term contracts, meaning future revenue from them could be volatile, drying up or even disappearing.
Anthropic estimated about 7% if its revue in the first nine months of 2026 came from customers it later found were violating its usage policy or commercial terms. The filing doesn’t specify the violation.
The company said it stepped up enforcement in August and September, which “may reduce our overall revenue and growth rates in future periods.”
Risk factors
Risk factors take up 94 pages of the filing—nearly a quarter of the document.
Anthropic wrote that models, “including our own, have exhibited, and may in the future exhibit, unexpected strategic or self-preserving behaviors, including attempts to resist shutdown.”
It also cited outputs that include “behavior resembling blackmails or efforts to influence human decision-makers to preserve the model’s continued operation or access.”
The company warned that insurance may not cover the fallout.
“With the scale of harm that could result from a catastrophic AI-related incident, no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure,” the company wrote.
Google and Amazon stand to benefit
Google and Amazon will collect the largest share of Anthropic’s compute spending. The filing shows Anthropic owes Google at least $114.5 billion through 2033 and Amazon at least $110 billion through 2036, and it must pay both “regardless of actual usage.”
Google’s take could grow further. Anthropic expects to spend another $125 billion to $150 billion on Google’s TPU capacity between 2028 and 2031.
