AI adoption is blocked by risk, not capability. Standards and insurance are

The thesis. AI adoption is not held back by model capability. It is held back by risk. As Rune Kvist, founder of AI Underwriting Company (AIU), explains, the binding constraint on deploying frontier AI in enterprises, banks, hospitals, and governments is the inability to make credible promises about what these systems will and will not do. Waymo illustrated this years ago: superhuman driving skill meant nothing without a trust framework that allowed regulators and insurers to say yes.

The blueprint. Kvist draws a direct line from electricity, cars, and nuclear energy to AI. Every transformative technology needed standards and insurance before mass adoption. Standards define the rules of the road and the tests that quantify risk. Insurers then price that risk and put capital behind it, creating a profit incentive to measure truthfully and reduce harm. AIU's AU1 standard for agent security, safety, and reliability is the first commercial attempt to build this layer for AI agents.

The standard. AU1 covers six risk categories, from hallucinations to jailbreaks to data leakage. It is updated quarterly through a consortium of Fortune 1000 risk leaders. Companies like Cursor, Harvey, Lovable, and 11 Labs have gone through the certification process, which includes third-party technical testing, policy controls, and independent audits by firms like KPMG. The output is a 100-page report that enterprises can trust because it comes from a neutral party, not from the vendor themselves.

The insurance link. AIU does not just certify; it underwrites. By partnering with Lloyd's of London, it turns a passing certification into an actual insurance policy, as it did for 11 Labs. This is the mechanism that creates trust: a conservative, centuries-old institution has looked at the data and is willing to put its balance sheet behind it.

The model layer. Beyond agents, Kvist sees the same gap opening at the model level, especially after the Fable and Mythos incidents. Governments need a neutral third party to run frontier technical evals and produce consistent, legible reports, just as Moody's does for bonds. AIU plans to expand from agent certification into model certification, and eventually into physical AI and robotics, where the liability stakes are even higher.

The tradeoff. A for-profit standard setter creates obvious tension: the incentive to hollow out standards for competitive advantage. Kvist's answer is that insurers act as a counterweight because they bear the financial consequences of weak standards. Whether that holds in practice remains unproven. There have been no claims yet, and the first major incident will test whether the framework is as robust as advertised. The skepticism is warranted, but the problem AIU is solving is real and growing fast.