Artificial intelligence (AI) company Anthropic reported a net loss of $80.7 billion in the first half of this year, according to a report by The Wall Street Journal (WSJ).

Citing a draft initial public offering (IPO) prospectus, the report said accounting charges linked to changes in the value of convertible notes and other fundraising instruments accounted for the vast majority of the loss.

Despite the reported loss, Anthropic generated $16.3 billion in revenue during the same period. However, the WSJ reported that just two customers contributed around one-fifth of the company’s sales.

Anthropic’s $80.7 billion loss marks a sharp increase from the $13 billion net loss it recorded during the first half of the previous year, the report said. The company did not immediately respond to requests for comment. Reports also suggest that Anthropic is targeting a stock market debut in November.

The company also faces challenges in securing sufficient cloud computing capacity from major technology providers, including Amazon Web Services (AWS), Google and Microsoft. According to the WSJ, these technology giants increasingly view Anthropic, the developer of Claude AI, as a competitor rather than simply a customer.

To address its computing requirements, Anthropic is reportedly leasing its own data centres and partnering with third-party data centre operators.

Meanwhile, the company has faced scrutiny over unintended actions by its AI models. Earlier this month, the US-based firm disclosed several incidents in which Claude carried out unintended actions involving external organisations, including websites operated by US government agencies. The disclosures prompted a White House directive calling on AI companies to strengthen their systems’ security.

In a report detailing previously undisclosed incidents, Anthropic identified four categories of unintended model behaviour. These included exploiting basic software vulnerabilities to execute commands, submitting forms without authorisation, and bypassing restrictions to access certain publicly available data.

Following these incidents, Anthropic restricted certain types of internet access for its AI models during the training and testing process.

The company said some incidents involved websites operated by federal, state and local government agencies. However, Anthropic declined to identify the affected organisations, citing requests from some of the parties involved.