Anthropic Warns AI Liability Could Exceed Insurance Capacity

Artificial intelligence company Anthropic has warned that a major AI-related disaster could lead to claims for money that are too large for any one insurance company to pay. The total cost could even be more than the entire insurance industry can cover.

In a registration statement filed on October 26 with the Securities and Exchange Commission for its planned initial public offering, Anthropic described major risks related to insurance and legal responsibility for its AI models.

The company said current insurance policies and contract protections might not cover losses caused by harmful AI behavior, especially when an incident causes widespread damage.

AI Insurance Market May Not Have Enough Coverage

Anthropic said insurance for risks specifically linked to AI is still a new and developing market. Because of this, the company may find it difficult to get insurance for some AI-related claims at prices it can reasonably afford. It may also be unable to get some types of coverage at all.

The company also questioned whether insurance companies have enough knowledge to judge these risks and enough money to cover a major AI disaster.

According to the filing, one incident or several related incidents could cause unusually large amounts of harm and lead to very costly claims.

Anthropic pointed to possible risks such as other people misusing its AI models, models behaving in unexpected ways, and AI being used in sensitive settings or situations where mistakes could cause serious harm.

The laws that decide who is responsible when AI causes harm are also still unclear. Anthropic said lawsuits could involve product liability, consumer protection, and other legal claims as courts and regulators work out how to handle AI-related harm.

Contracts Could Create More Legal and Financial Risks

Anthropic acknowledged that its contract protections might not be enough to limit its legal responsibility under the laws that apply.

The company has also agreed to take on extra responsibility through contracts with outside experts and testing companies that assess its advanced AI models.

These contracts include promises to cover certain losses caused by security incidents, data breaches, lost data, and other harms linked to testing.

Most of these agreements limit how much Anthropic must pay. However, some contracts require the company to defend outside evaluators and cover their costs for certain claims made by third parties, with no limit on the amount it may have to pay.

These claims could include accusations that AI models or their outputs violate intellectual property rights or other ownership rights.

Anthropic said unexpected behavior by its models could lead to large claims, even when users or developers did not intend for the models to take those actions.

Cybersecurity and Insurance for Company Leaders

Anthropic also pointed out limits in insurance coverage for cybersecurity incidents.

The company has insurance policies that may cover some losses related to security problems. However, it warned that its current coverage might not be enough and that insurers might not pay claims in full or on time.

Anthropic also discussed the possibility that insurance for company directors and officers could become more expensive as it becomes a publicly traded company.

The filing said changes in regulations and rules for how companies are run could raise insurance costs.

As a result, Anthropic might have to accept lower coverage limits, less protection, or much higher premiums to keep a similar level of insurance.

Past Lawsuits Show That Insurance May Not Cover All Costs

Anthropic also referred to its experience with copyright lawsuits, including the Bartz case.

The company disclosed that its insurance did not cover all the costs of that lawsuit.

It warned that future claims could also cost more than its insurance covers, leaving the company to pay additional expenses itself.

Anthropic also pointed to possible legal costs and duties to cover losses involving business partners and current or former company directors and officers.