An amended IPO filing details Anthropic’s growing reliance on Claude for coding, research and operations, while distinguishing usage valued at customer prices from actual spending and acknowledging safety risks as AI helps develop future models.
Anthropic’s internal use of Claude reached a valuation above its monthly personnel expenses in April 2026, according to an amended initial public offering filing, illustrating internal adoption.
The comparison values Claude usage at the prices Anthropic charges customers and excludes stock-based compensation from employee expenses. It therefore does not establish that the company’s actual spending on internal AI exceeded its payroll.
In the filing with the US Securities and Exchange Commission, Anthropic said the value of its internal Claude usage increased more than 250 times between January and September 2026. The disclosures describe a growing role for AI across model development and business functions.
Claude takes larger role in development
More than 90 per cent of contributions to Anthropic’s codebase in September were written by Claude, the company said. During the three months ended September 30, average daily code contributions per engineer exceeded the corresponding 2024 average by more than 45 times.
Anthropic said engineers increasingly direct and review work produced by AI rather than writing software line by line. Claude is also used to conduct experiments, create and process training data, automate evaluations and safety audits, and write and review code.
The company described successive models as contributing to the development of their replacements. In its account, the computing resources assigned to internal Claude use support a feedback loop in which better tools help employees build subsequent systems more quickly.
The filing also contrasted workforce expansion with revenue growth. Anthropic’s headcount increased more than fivefold between December 31, 2024, and September 30, 2026, while revenue in the July-September 2026 quarter was 86 times the level recorded in the October-December 2024 quarter.
Anthropic attributed part of that difference to Claude’s deployment across research, engineering, sales and operations. Employees can coordinate multiple AI agents to carry out tasks that would otherwise require larger teams, according to the company.
Those comparisons reflect Anthropic’s assessment of its operating leverage. The code-volume figures measure the amount contributed, rather than independently demonstrating equivalent improvements in software quality or the value delivered by each contribution.
Faster model development raises safety questions
Alongside its productivity claims, Anthropic acknowledged risks from using AI to develop increasingly capable models. It said the approach could present considerable safety concerns, particularly if it speeds up technological progress.
The company pointed to its Responsible Scaling Policy as a framework for monitoring and mitigating risks associated with recursive self-improvement, in which AI systems contribute to creating more advanced versions of themselves.
Anthropic said it believes investment in safety techniques will allow it to manage these risks while making greater use of Claude in model development. The filing thus presents internal AI deployment as both a source of operational capacity and an area requiring oversight as the technology takes a larger role in building future systems.
