AI Liability Now Outruns What Insurers Will Cover — And the Gap Is Yours
Insurers cap AI liability as low as $50M while one failure can cost billions. The uninsured gap sits with deployers — how to protect your business.
What happened
Underwriters including Munich Re are now capping AI-related liability coverage as low as $50 million — while the exposure from a single widely deployed failure runs into the tens or hundreds of billions, according to industry reporting. The gap between what insurers will cover and what a failure can cost has become a chasm.
The problem is structural. Insurance depends on uncorrelated risk: houses do not all burn down on the same night. Model failures are correlated by construction — if one widely deployed model has a flaw, every customer has the same flaw at the same time. That is the risk shape insurance handles worst. And there is no actuarial history to price from, because frontier models change faster than loss data can accumulate.
Why this matters
The uninsured remainder sits with whoever deployed the system. As AI agents take more external actions — sending messages, moving money, changing systems — the surface area for correlated failures grows, and the insurance industry is responding by limiting its own exposure first.
This arrives in the same week that U.S. states are suing over alleged rogue agents and companies are reporting agents taking unintended actions. The legal and financial risks of deploying AI are compounding while the insurance backstop is shrinking.
What this means for your business
1. Know your actual coverage. Review your cyber and professional liability policies for AI exclusions and sub-limits. Many policies quietly cap AI-related events far below your general limits — find out before an incident, not during one.
2. Treat the uninsured remainder as your problem. If your insurer caps AI liability at $50 million and your deployed system touches millions of customers, the gap is yours. Size it honestly in your risk planning.
3. Reduce correlated blast radius. Limit how widely a single model or agent configuration is deployed without variation. Staged rollouts, canary deployments, and diversified model providers are not just engineering practices — they are financial risk controls.
4. Invest in prevention, not just coverage. Since the insurance backstop is thin, the return on safety work is higher: testing, approval gates, audit logs, and kill switches directly reduce the exposure you are self-insuring.
5. Put AI liability in writing with vendors. Your contracts should say who bears the cost of a model failure — you or the provider. In a correlated-failure scenario, ambiguous contracts become expensive disputes. Clarify liability, indemnification, and insurance requirements before deployment, not after.