Anthropic reports 5% to 7% of revenue from policy violators

The scope of policy violations

Anthropic recently filed an S-1 registration statement with the SEC, revealing a specific breakdown of its revenue streams. The company estimates that between 5% and 7% of its earnings during the first three quarters of 2026 originated from users who broke its commercial rules or usage guidelines. This data provides a rare look at the scale of non-compliant activity on the platform.

The filing explains that these violations are identified through a detection and review process. Because investigating and enforcing these rules takes time, Anthropic continues to record revenue while the services are still being provided. The company only restricts or terminates access once the violation is confirmed and processed.

Anthropic notes that this estimate requires significant judgment from the company. The figures may change as the company develops better detection methodologies and identifies usage that was previously hidden. The company also stated that this dynamic is common in the broader AI industry.

Financial impact and scale

While the S-1 does not provide a specific dollar amount for the nine-month period, the scale of the violation is notable. Based on reported revenue of $16,286 million for the first six months of 2026, the revenue from violators is estimated to be no less than roughly $814 million to $1.14 billion. This calculation assumes the revenue for the final quarter of the period was at least equal to the previous half.

The company's quarterly revenue figures show a jump from $4,731 million in the March quarter to $11,555 million in the June quarter. These figures help contextualize the growth the company is experiencing while simultaneously trying to police its user base. The exact dollar amount of the violations depends on the September-quarter revenue, which was omitted from the filing.

The disclosure is intended to inform potential investors about the factors affecting the company's results. It highlights a specific risk where enforcement actions could lead to variability in results from period to period. By being transparent about these numbers, Anthropic is acknowledging the friction between scaling a product and maintaining strict governance.

The challenge of enforcement

Anthropic has indicated that it is actively working to reduce the share of revenue coming from these violators. The company expects the percentage to decline in future periods as it enhances its detection and enforcement efforts. These efforts were specifically stepped up during the months of August and September 2026.

One of the primary hurdles for the company is the persistence of non-compliant users. The S-1 warns that customers who are restricted or terminated may simply try to re-register under new accounts. This creates a continuous cycle of detection and evasion that requires constant technical updates to the platform's permission layers.

The company admits that the timing and magnitude of any adverse effects on revenue growth remain uncertain. While they want to reduce the percentage of violators, doing so might also impact the overall growth rate of the business. This balance between safety, compliance, and commercial expansion is a central challenge for the company as it moves toward a public offering.

Context on data distillation

The broader landscape of policy violations includes concerns over data distillation, where models are trained using the outputs of other AI systems. Anthropic has previously shared evidence suggesting that specific methods have been successful in addressing this issue. According to the company, these defenses have cut the volume of data collected by distillers by a margin of roughly 90 to 99%.

Reports from The Information, as relayed by Reuters, indicate that other major players like DeepSeek and Moonshot AI used a significant amount of data from OpenAI’s GPT-6 Astra to build their own latest models. This context highlights the ongoing struggle for AI labs to protect their intellectual property and maintain the integrity of their commercial terms against unauthorized data harvesting.