AICPA and AIUC combine security audits with live AI agent testing

The American Institute of CPAs and Artificial Intelligence Underwriting Company introduced a joint assurance offering on October 19, 2026, pairing an examination of business systems with live testing of AI agents. The CPA-led engagement combines a SOC 2 report with AIUC-1 certification to assess both system controls and whether agents behave as intended.

Several CPA firms are already delivering the combined service to clients, the partners said. A full launch of the joint certification is expected in early 2027, after the initial engagements conclude. That puts the announcement at the start of delivery, rather than at the completion of the rollout.

Two examinations in one engagement

SOC 2, short for System and Organization Controls, is an independent examination performed by a licensed CPA under AICPA standards. It evaluates controls relevant to a system’s security, availability, processing integrity, confidentiality or privacy. The resulting report includes an opinion on how those controls are designed and how effectively they operate.

AIUC-1 supplies the agent-focused part of the engagement. Its controls cover six principles: data and privacy, security, safety, reliability, accountability, and society. According to the partners, the standard was developed by a consortium of more than 250 chief information security officers and security leaders. It is updated quarterly to reflect changing AI risks.

The AIUC-1 controls also map to frameworks including MITRE ATLAS, ISO 42001 and the EU AI Act. The joint offering brings the SOC 2 examination and AIUC-1 certification into a single report delivered through one CPA-led engagement, rather than presenting them as two separate purchasing decisions.

Testing behavior, not just policies

AIUC cofounder Rajiv Dattani describes live behavioral testing as the distinction between AIUC-1 and a review limited to policies and guardrails. He says agents face hundreds of risk and attack combinations. Each combination is evaluated through iterative conversations, with a typical evaluation covering thousands of scenarios.

• Hallucinations and data leakage are among the failure modes included in the testing.
• Manipulative requests test agent behavior when an interaction attempts to steer it in an unwanted direction.
• Adversarial prompt injection—attempts to redirect an agent through hostile instructions—is another named attack category.

The partners tie those tests to the access agents receive inside businesses. Their announcement points to agents holding credentials, calling software interfaces and acting inside core systems. Their stated goal is assurance that both an agent and the systems it touches are secure, while checking that its behavior meets enterprise requirements.

Companies have made substantial investments in AI and investors, clients and other stakeholders deserve a fuller, objective picture of whether the technology is safe, reliable and properly governed.

Mark Koziel, AICPA president and CEO, in the joint announcement

Assurance that must be renewed

The offering is not structured as a one-time examination. Both SOC 2 reports and AIUC-1 certifications require annual renewal, and AIUC-1 adds quarterly technical retesting. The partners say that recurring work is intended to verify that an organization continues to meet evolving standards for trust, security and reliability.

The commercial aim is to reduce the repeated security questionnaires that accompany enterprise AI buying. AICPA and AIUC describe those exchanges as slow, costly and often inconclusive. They are positioning the combined report as a common reference for buyers assessing AI business risk, rather than negotiating trust through questionnaires alone.