Alex Lazarow is an author, speaker, global venture capitalist and contributor to The FR.

Lloyd’s of London is over 300 years old. It was built to insure ships in oceans no one trusted. The Lloyd’s of AI hasn’t been built yet — but it’s coming.

Trust institutions rarely show up first. They arrive late, after the technology has already broken something.

For example, the Big Four accounting firms were all founded around 1850. That timing is not an accident: the industrial revolution had produced firms too large and too complex for owners to verify directly, and a new profession emerged to do the verifying. Independent auditors became the trust layer that made modern capital markets possible.

Underwriters Laboratories emerged because electrification was powerful, useful, and occasionally catastrophic. The certificate authorities that quietly run SSL only existed because e-commerce demanded a way to know the site asking for your credit card was real. And of course, Lloyd’s of London began as a coffeehouse where shipowners tried to price risks they could not see.

AI is now at a similar point: adoption is accelerating, but the trust institutions needed to manage the technology’s risks have yet to catch up.

To remedy it, a new verification layer is emerging, and I think it’s the most important venture category nobody is talking about yet.

The early wave verified what AI said

An early wave of AI verification focused on confirming models were safe and accurate and that their claims were real. From late 2022 through roughly the end of 2024, these tools addressed a discrete set of problems facing AI models at the time.

That included some degree of hallucination detection, prompt-injection defense and deepfake detection.

Implicit in the first wave was the assumption that a human or deterministic system would read the output and act on it. The verification layer only needed to flag problematic outputs.

Agents are changing everything.

The next wave verifies what AI does

In 2025, AI stopped producing text for humans to read and started doing work for itself. Trades were booked. Refunds were processed. Articles were filed. Code shipped to production. The verification question changed shape.

The question is no longer, “Is this output accurate?” It is, “Can we trust the action AI took?”

It is now, “Can someone stand behind the work that just happened?”

The most widely adopted use case of AI is arguably coding. Unsurprisingly, this has been a big area of investment for verification. But the use cases are expanding, and so too are the verification styles of companies, including in fintech and insuretech.

The Artificial Intelligence Underwriting Company launched in July 2025, among a few startups in this category. Its product is insurance for AI agents, priced against a new audit standard called AIUC-1, plus an underwriting process that ties premium to evidence.

Objection is building a journalist ranking and verification network to “adjudicate the truth of journalism. If the next decade of frontier AI is trained on the public corpus of news, verifying news verifies the foundation.

Oath, a Fluent incubation, recently launched as a licensed audit firm built specifically for AI-generated financial work. The thesis: as AI increasingly completes bookkeeping and tax preparation, someone still has to sign off on the financials.

At the end of the day, each is built on the same idea: take a verification function that once belonged to a human profession and rebuild it for a world in which the underlying work is autonomous.

The verification surface area is Increasing

The obvious objection: if AI does the verification, the verifiers disappear. We believe the historical record argues the opposite.

Start with developers. ChatGPT shipped in November 2022. Since then, AI coding tools have moved from novelty to default. A March 2026 Boston University report finds 84% of developers now use or plan to use them. Over the same period, U.S. software developer employment hit a record 2.5 million, up more than 400,000 since 2022. Headcount went up, not down.

Accountants tell the same story over a longer arc. The profession absorbed mainframes, ERP systems, the cloud, and now AI. The Bureau of Labor Statistics counted over 1.5 million accountants and auditors in 2024, a record, and projects above-average growth over the next decade. BLS’s own framing: automation of routine tasks is expected to increase demand for advisory and analytical work, not reduce it.

This is Jevons’ paradox: Cut the unit cost of something useful, and demand tends to grow faster than the cost falls. Benedict Evans made a similar argument about AI’s exposure to jobs: Make something valuable cheaper, and the world finds more uses for it.

Verification is in that bucket.

Where are we heading?

The first wave verified what AI said. The second wave is verifying what AI did.

Enjoying? Sign up here

The space remains highly dynamic. Standards will emerge, including some that machines can issue to other machines, backed by a human institution willing to assume legal liability when the chain breaks. We are not there yet.

I also expect a wave of players that guarantee outcomes or assume the risk of failure, much as Lloyd’s insures against other risks.