Anthropic reported that 5% to 7% of its revenue during the first nine months of 2026 originated from users violating its terms of service. The company identified Chinese AI labs, including DeepSeek and Moonshot, as the primary violators using model distillation to train competing systems.
In February 2026, these labs utilized approximately 24,000 fake accounts to perform over 16 million interactions with the Claude models. Anthropic stated that these accounts paid for access, which allowed the unauthorized data extraction to be recorded as legitimate revenue on the company's income statement.
The company implemented stricter security measures, blocking tens of thousands of accounts. By the third quarter of 2026, the activity was largely halted, and new security protocols deployed in October 2026 reportedly stopped 90% to 99% of unauthorized data harvesting.
Anthropic reported third-quarter revenue of $16.8 billion, compared to $11.5 billion in the second quarter. Total revenue for the first half of 2026 reached approximately $16.3 billion. The company faces risks from high customer concentration, with two unnamed clients accounting for 20% of first-half revenue.
Ahead of a planned IPO, investor discussions suggest the company's valuation could be reduced from an initial target of $2 trillion to $1.5 trillion due to slowing growth. The firm also disclosed significant long-term financial obligations related to computing resources.
