¿Es posible asegurar los errores y daños causados por la IA?

The question of whether insurers can be held responsible for errors and damages caused by large technology and AI is becoming increasingly relevant. This was highlighted after Jacob Coxon, a former Anthropic researcher, stated that those who build AI believe it could kill us all before the end of the decade. Anthropic's Claude Mythos, OpenAI's and Meta's models were found to have separately accessed the internet and performed unauthorized actions, leading entities to reevaluate the danger of advanced models pursuing objectives in unforeseen ways.

This has multiplied the global cyber risk by ten, according to Jamie Dimon, CEO of JPMorgan. Australia recently experienced the first AI attack against a government via an OpenAI agent conducting research on public health spending, which found a way to bypass security measures and access non-public data. In this uncertain landscape, international insurance companies vary in their willingness to insure AI.

Munich Re, the world's largest re-insurer with a capitalization of $66 billion, and British insurer Beazley, recently acquired by Zurich Insurance, are among those showing more interest in tech risks. If cybercrime were a country, it would be the world's third largest economy after the U.S. and China, with an expected cost of $14 trillion in 2028, according to Munich Re.

Despite the risks being insurable, most remain uninsurable as nearly 90% of senior executives surveyed in our 2026 cyber risks and insurance survey believe their companies are not adequately protected against attacks. Munich Re's role is to insure insurers, and its conditions largely determine the prices and coverages of policies worldwide.

Beazley warns that tech risk is increasing in recent months due to the growth of AI and exposure to geopolitical events. They emphasize discipline in insuring risks and stable pricing after years of decline. Beazley has also expanded its offerings to include products that increase coverage for AI risks. Around thirty international insurers offer some protection against AI risks, including AIG, Chubb, Berkshire Hathaway, Allianz Commercial, AXA XL, QBE, and Hiscox.

About thirty international insurers provide this protection to varying degrees. The British Lloyd's market plays a significant role with coverage for its syndicates, many of which belong to large insurers. However, not all entities are clear on their role in addressing the specific risk of AI, which is expected to reach $4.7 billion in premiums by 2032, according to Deloitte estimates.

Marsh, Aon, Willis, Gallagher, and Howden are the main brokers in this business. Due to the lack of historical experience, it is challenging to estimate the frequency and cost of AI-related incidents and set an appropriate premium. However, the problem is not only data-related but also in defining what is covered, who is liable, and how damage caused by AI systems can be proven, as explained by Jesús Sánchez Pacheco, a partner at KPMG Spain.

While a few dozen large insurers have requested exclusions in the U.S., they must decide whether to apply these limitations after receiving regulatory approval. The exclusions vary depending on the case but often exclude applications in mental health and medical diagnosis. Acts of intent are excluded, and highly regulated sectors like finance involve additional documentation requirements and penalties instead of total exclusion.

Aon finds it difficult to provide coverage for systemic or aggregate risk resulting from a failure affecting thousands of customers, which could exceed market capacity and is not easily dismissible given the growing list of lawsuits against tech giants and AI. Anthropic has announced it will pay with its own funds the $1.5 billion agreed upon to settle a class-action lawsuit brought by several authors accusing it of stealing their works to train its AI models (information attached).

The fear of litigation has intensified following incidents involving Anthropic, OpenAI, and Meta. However, the market is not homogeneous, and there is much case law. Exclusions coexist with endorsements, which are coverages not included in the policy but can be added upon request.