Anthropic Warns of “Catastrophic or Existential Risks To Humanity” Ahead of IPO

Anthropic's filing for a major initial public offering (IPO), which is when a company first sells shares to the public, included nearly 100 pages of risks. This was an unusually detailed list of ways its artificial intelligence business could go wrong.

The list covered everything from slower revenue growth and government rules to “catastrophic or existential risks to humanity.” The wide range of risks shows the unique challenge investors face when judging both the promise and the dangers of Anthropic's technology.

Here are some of the most important points from the filing:

Catastrophic harms

Anthropic pointed out that its AI could “behave in unintended, misaligned, or autonomous ways.” The company said its technology could threaten humanity if it was developed without enough safety measures and human oversight. This could be especially dangerous if AI systems were misused, copied, turned into weapons, or allowed to act largely on their own.

The company warned that these risks grow as its AI models become more able to act independently. These abilities could “increase the potential for harm,” including “consequences that may be difficult or impossible to reverse.” Examples include changing computer code, deleting or changing data, and carrying out financial transactions.

The filing warned that human reviewers might not be able to stop harmful actions. Anthropic also said its technical systems for monitoring AI could have limits if its models found ways to “circumvent such techniques” by hiding what they were trying to do.

As Anthropic begins using its AI for scientific work, it also warned that the technology could help release dangerous substances. The company said that “such consequences may be immediate and irreversible.”

Insurance problems

Anthropic also warned that existing insurance policies “may prove insufficient” to protect AI companies from legal responsibility when their technology causes real-world harm. This could be especially serious after a “catastrophic AI-related incident.”

“No single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure,” the company wrote. In other words, one insurer, or even the insurance industry as a whole, might not have enough money or knowledge to provide coverage for risks of this size.

Slower revenue growth

Anthropic's revenue grew by more than 1,000% year over year in 2025. However, the company warned that it “does not expect” revenue to keep growing at similar rates. It pointed to a more mature AI market, stronger competition, and safety concerns that could delay the release of its models as possible reasons for slower growth. The company also said it had “struggled” to predict its revenue accurately in the past and might face the same problem again because the AI market is still new.

Bloomberg News previously reported that Anthropic's growth from the second quarter to the third quarter of 2026 slowed to about 45%. In the previous quarter, growth had been more than 140%.

Computing supply

Another major risk is that Anthropic might not be able to get enough computing power, or that the cost of this power could rise “significantly.” Some of its current cloud computing partners also compete with Anthropic. Because of this, these providers “may not always be fully aligned” with the “long-term interests” of the company that makes Claude, the filing said. Anthropic has major agreements to use chips, data center space, and other computing equipment from companies such as Alphabet Inc.'s Google, Amazon.com Inc., and Elon Musk's SpaceXAI. These companies are also developing their own AI models.

Anthropic also warned that spending too much based on incorrect estimates of its needs could leave it with more computing capacity than it can use, wasting money. “Our actual compute requirements depend on numerous factors that are difficult to predict,” the company said. These factors include how quickly its research advances and how much computing power its training methods require.

Government scrutiny

Anthropic warned that governments around the world could restrict its technology, putting an important part of its business at risk. The company said that “governmental authorities have restricted, and may in the future further restrict, the availability of our models.”

In June, the US government imposed export controls on Anthropic's Fable 5 model because of security concerns. The company has also had a difficult relationship with the Trump administration. It sued the Department of Defense after Defense Secretary Pete Hegseth moved in March to label Anthropic a supply-chain risk, following a dispute over surveillance and autonomous weapons.

Opposition to data centers

Anthropic sees growing public opposition to the huge data centers it needs to train and run its software as a threat to its business. In its filing, the company said this opposition had “become more organized and coordinated.” Anthropic said most of its data centers are currently in the US. It warned that if a major facility were lost or badly disrupted, its other existing or future facilities might not be able to make up for the loss quickly.

Anthropic said it had voluntarily promised to pay for some of the effects of its data centers' energy use, such as improvements to the electricity grid. However, the company warned that these commitments could raise its costs if they became standard practice.

Hackers

Anthropic sees cybersecurity, or protecting computer systems from attacks, as an important business opportunity and a risk to its own operations.

The AI company said it faces several security threats. Hackers could attack Anthropic itself or use its tools to attack others, something the company said is already happening. Anthropic also expects to become a more frequent target of cyberespionage backed by governments, especially attacks aimed at its most advanced AI models.

TSMC

Anthropic pointed out that most of the world's most advanced computer chips are still made in Taiwan, mainly by Taiwan Semiconductor Manufacturing Co. This creates major risks, including possible disruptions caused by political and military tensions between Taiwan and China, the AI company said.

Company culture

Over the years, Anthropic has built a workplace culture that it describes as a “low-ego, truth-seeking environment.” The company also said many of its key employees are “strongly motivated by our mission and the opportunity to work on addressing specific problems within the AI landscape.”

Anthropic warned that important employees might leave for other jobs they believe better match their interests as the company is required to “scale and mature as an organization.” In short, Anthropic expects to face growing pains as it grows and becomes a more established company.