Anthropic’s IPO filing reveals that over 90% of its code is generated by its AI model Claude, which significantly boosts productivity but raises risks of rapid, potentially unsafe AI development and propagation of model flaws. The company highlights concerns about AI-driven scientific research, including physical safety risks and misuse by malicious actors, and advocates for industry-wide regulation and independent safety reviews to mitigate these challenges.
Anthropic, an AI company, recently filed for an initial public offering (IPO), providing public investors with insight into its operations and financials. The filing revealed that over 90% of Anthropic’s code was written by its own AI model, Claude, which plays a central role in the company’s research, design, and development processes. Claude automates experiments, generates training data, conducts safety audits, and improves algorithms, significantly boosting the company’s efficiency. Financially, Anthropic reported $16.3 billion in revenue and $693 million in operating losses in the first half of 2026, but achieved profitability in Q2 2026 with $974 million in operating income.
The company heavily relies on Claude, with internal AI usage costs now exceeding salary expenses. Claude has dramatically increased engineering productivity, enabling a 45-fold increase in code contributions per day compared to 2024. However, this rapid self-improvement and accelerated AI development pose significant risks. Anthropic warns that the fast pace could outstrip the company’s ability to implement adequate safety and security measures, and flaws in current models might propagate into future versions, complicating detection and correction.
To mitigate these risks, Anthropic advocates for an industry-wide initiative to regulate AI development pace and enable independent safety reviews of frontier models. The company plans to invest $1 billion over five years to build external reviewer capacity in partnership with Accenture. This reflects a broader concern about the challenges of ensuring AI safety as models become increasingly autonomous in their development.
Anthropic also emphasizes its commitment to using AI for scientific research, highlighting breakthroughs made with Claude, such as discovering a new enzyme family relevant to gene editing. The company has developed AI-powered tools like Claude Science and Claude for Life Sciences to assist in drug discovery and regulatory processes. Additionally, Anthropic has created software that enables AI agents to operate laboratory equipment, automating research tasks and collaborating with major pharmaceutical and research institutions.
Despite these advancements, Anthropic acknowledges significant risks associated with AI’s role in scientific research. The ability of AI models to control lab equipment could lead to physical harm, property damage, or hazardous releases due to misalignments or misuse. Furthermore, the company warns that its models could be exploited by malicious actors to develop dangerous weapons, despite existing technical safeguards. These safeguards may be incomplete or circumvented, underscoring the ongoing challenges in managing AI safety and security.