Anthropic S-1 filing discloses cyber attack and AI safeguard limits as risks

Anthropic submitted an S-1 to list on American stock exchanges, releasing its first public and audited financial disclosures on Monday, five years after the company started in 2021.

The prospectus contains 92 pages of risks to the business, spanning competitive, regulatory, operational, and existential threats. The filing marks the company’s first public look at its audited finances. The risk section takes up cyber attacks, model safeguards, liability, infrastructure concentration, and immigration policy.

Anthropic said it identified what it considers the first recorded major cyber attack carried out with minimal human involvement. The firm attributes the intrusion attempt to a Chinese state-backed hacking group that used a Claude coding tool against about 30 targets worldwide. The intended victims covered major technology firms, banks, chemical manufacturers, and government bodies. This detail comes from Anthropic’s disclosure of the suspected Chinese state-sponsored intrusion.

Anthropic said its AI safeguards may not work because they are incomplete, can become outdated, have been circumvented before, or may fail to operate as intended. The company disclosed that its technical guardrails may not block misuse as threats evolve.

Its strongest models could help bad actors create chemical, biological, or other dangerous weapons. The filing warns that frontier models can provide meaningful assistance to malicious actors seeking weapons capabilities. Company executives wrote in the S-1 that frontier systems could provide “meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons”.

The filing organizes dangers AI poses to society and human beings into four groups: authoritarian abuse, economic disruption, loss of control, and misuse by bad actors. Anthropic warned that authoritarian governments may use AI to deepen repression and build high-tech surveillance states. The company said it would work with free societies to counter authoritarian abuse and promote freedom and democracy.

Anthropic flagged AI-enabled catastrophic events as a risk to the business, including doomsday-style scenarios that employees have warned about. The company warned of public relations harm if its models are used to create bioweapons or other unintended outputs.

Anthropic fell out with the Trump administration after refusing to change model restrictions for domestic surveillance or autonomous lethal weapons. The disagreement centered on whether Claude could be used for domestic surveillance or autonomous lethal weapons. The company is in a standoff with the administration over restrictions on Department of Defense use of Claude.

Anthropic said it faces liability exposure because AI insurance options are still evolving. The S-1 warns the company might not get insurance for some AI-related liabilities at affordable prices. The filing states the company may be unable to get coverage on commercially reasonable terms or at all, leaving it open to large uninsured losses.

Anthropic reported $16.3bn in revenue and an $80bn loss for the first half of 2026. A $72.4bn one-off charge drove much of that loss, and the filing attributes most of the net loss to this single charge. In 2025, Anthropic brought in $4.585bn in revenue and lost roughly $33.9bn. These figures come from the audited financials in the S-1.

The company says it has seen explosive growth in every year of its existence.

Most of Anthropic’s spending goes toward obtaining computing power. It laid out almost $9bn for AI infrastructure in the first half of 2026, compared with $2.3bn during the same months in 2025. The filing notes that AI infrastructure spending nearly quadrupled year over year in the first half.

The firm’s top five customers represent 25% of its business. Losing any top buyer could deal Anthropic a serious blow because of its enterprise focus, with customer concentration among a few enterprise clients heightening this exposure. Anthropic said its customers are mainly other businesses rather than consumers, contrasting its enterprise focus with OpenAI’s approach. Executives wrote that frontier AI creates the greatest economic impact when deployed across enterprises, where real-world deployment is where frontier AI is most valuable.

As a public benefit corporation, Anthropic must balance profit with social goals, which may hinder short- or medium-term financial aims. The firm must report public benefit progress in the same manner as its financial results.

The company named Google and OpenAI as its main competitors in a highly competitive space. Anthropic said its success depends partly on whether top executives remain at the firm. CEO Dario Amodei has called for a pause in AI development to encourage responsible practices and prevent catastrophe.

Anthropic cited the uncertain regulatory landscape for AI and datacenters as a risk. Because its AI infrastructure sits in the United States, the company faces regulatory shifts, electricity grid limits, natural disasters, and resistance to AI or datacenters. The filing specifically raised concern about datacenter moratoriums that have passed or are under consideration. Anthropic is expanding data center capacity internationally to diversify its footprint and reduce concentration risks, aiming to lessen reliance on US-based infrastructure.

The company said talent is a major part of its value proposition. It worries that US immigration policies could impair recruitment and retention of highly qualified employees. The Trump administration has targeted H-1B visas, which technology firms use to hire global tech and engineering talent.

Anthropic described AI potential as millions of virtual agents able to transform many fields, a concept the company summed up as a country of geniuses in a datacenter. The filing listed software, physics, math, biology, medicine, finance, economics, commerce, robotics, military and intelligence capabilities as areas AI could transform.

Military and intelligence capabilities appear in the list despite Anthropic’s standoff with the Trump administration. The company highlighted four opportunity areas: biology and medicine progress, economic growth, the global spread of growth, and streamlined government services such as social services and medical programming.

It said it works with some pharmaceutical and biotechnology companies and that Claude cut clinical trial report writing from months to minutes. The company pointed to clinical trials as an example of AI acceleration in biology and medicine. Executives claim Claude helped produce biological discoveries, including a molecular machine that could become a novel biotechnology tool for gene editing.