SOC 2 Plus AIUC-1: Would the AI Inside Your Firm Pass the Same Test?

This week brought good news for the profession. The AICPA and the Artificial Intelligence Underwriting Company (AIUC) announced a joint offering that pairs a SOC 2 report with AIUC-1 certification in a single CPA-led engagement. A practitioner can now attest that a client's AI agents were tested live by people trying to make them misbehave, and that they're retested every quarter. Several firms are already delivering it ahead of a full launch in early 2027.

Boards, investors and customers all want to know whether AI behaves the way it should, and they need someone independent to tell them. A CPA-led engagement brings the independence and professional judgment the trust economy runs on. Firms that build this capability early will have a new service to offer their clients, along with a new standard to think about closer to home.

The question that comes next

The same firms that will sign these reports are rolling out AI on their own engagements right now, some through vendors and some through tools they built in-house. So what would a SOC 2 report and AIUC-1 certification say about the AI that touches your own client evidence?

Here's what we've been hearing from firms over the last couple of months. Leaders are adding AI certifications to their vendor checklists. Security review boards at the large firms want to know how an agent behaves when someone tries to break it, and a vendor's SOC 2 doesn't answer that. Audit clients are starting to ask their auditors which AI touches their data and whether it gets used for training. Firms are under real pressure to have good answers, and most are still working on them.

Trust needs a retest cycle

Whether a firm's AI comes from a vendor or from its own team, the firm needs a way to know its agents have been tested and can be trusted, and for that to be verified at regular intervals. The standard updates every quarter, and the AI gets retested every quarter.

When firms build AI in-house, that recurring testing is the part that tends to get overlooked. I understand why firms build. A first version is cheap to build now, and practitioners know their methodology better than anyone. The retest cycle is a cost that doesn't show up until after the build is finished, and then it doesn't stop.

Why we sought it out

We chose to hold ourselves to this standard early. Fieldguide is SOC 2 compliant, and in May we became the first audit and advisory platform to earn AIUC-1 certification. Firms run their clients' most sensitive evidence through our platform, and we didn't think they should have to take our word for how it behaves. We've been on the quarterly retest cycle since.

Living on that cycle has shown us how much it takes to keep up. Even firms with strong engineering talent will find it hard to keep pace with where AI is going on their own, and the retest cycle adds to that. A partner who carries it gives practitioners a foundation that's already been tested, so the firm can spend its time building what makes it different.

Where to start

I'd start with a simple inventory of the AI that touches client evidence, whether it comes from a vendor or from inside the firm. Ask vendors how their agents have been tested beyond a SOC 2. For anything built in-house, decide who owns the retest and what it costs to keep going. If the answer is "we don't know yet," that's okay. Many firms are there.

The profession has always asked clients to show their work. I'm curious how firm leaders are thinking about holding their own AI to that same standard, and whether the AI you run, in-house or with a vendor, would pass the test you're about to start selling.