AICPA adds AI agent audits to SOC 2 with AIUC-1
AICPA and AIUC now combine a SOC 2 report with AIUC-1 AI agent certification in one CPA-led audit, with quarterly retests. What to ask your AI vendors.
The AICPA and the Artificial Intelligence Underwriting Company (AIUC) launched a combined AI agent audit: a SOC 2 report and an AIUC-1 certification in one CPA-led engagement. SOC 2 is the report your customers already ask for. AIUC-1 tests how an AI agent actually behaves. Putting them in one report means "is your AI safe?" is about to become a line on the same vendor questionnaire as "do you encrypt data at rest?"
What actually happened
Per the AICPA release and the Journal of Accountancy:
- The offering: One report from one CPA-led engagement, covering SOC 2 controls and AIUC-1 certification.
- What AIUC-1 tests: Agents are tested live, not just their written policies. That covers hundreds of risk and attack combinations, including hallucinations, data leakage, manipulative requests and prompt injection.
- Upkeep: Both need annual renewal. AIUC-1 also requires quarterly technical retesting.
- Timing: Several CPA firms are already doing it (none named). Full launch is expected in early 2027.
- Scale: The AICPA says more than 40,000 organizations got a SOC 2 attestation or an AIUC-1 certification last year. That figure counts both separately, not the new combined report.
No pricing was announced.
Why AI agent audits matter for your business
This cuts both ways for a small operator.
If you buy AI tools, you now have a standard question to ask: "Do you have a SOC 2 with AIUC-1, or a date for one?" A vendor whose agent reads your inbox, touches your store or talks to your customers should be able to answer. A vague "we take safety seriously" is an answer too.
If you sell anything with an agent in it, expect that question from bigger customers in 2027. Most small shops can't fund an audit cycle with quarterly retests. You can still be ready for one:
- Log every agent action with who or what triggered it.
- Scope permissions per task. An agent that drafts replies doesn't need send rights.
- Keep a prompt-injection test set and run it every time you change the model or the prompt.
Those three things are most of what an auditor will want to see. They also prevent the incidents that make audits matter.
Key takeaways
- AICPA and AIUC combine SOC 2 and AIUC-1 AI agent certification in one CPA-led report
- AIUC-1 tests live agent behavior: hallucinations, data leakage, prompt injection
- Annual renewal plus quarterly retesting; full launch expected early 2027; no price announced
- Add "SOC 2 with AIUC-1?" to your AI vendor questions now
- If you ship agents, start logging actions, scoping permissions and testing injections today
Building an agent you'll have to defend later? We build AI agents with action logs, scoped permissions and injection tests from day one — the things an auditor will ask for. See how we build, or talk to us about your agent.
Sources: AICPA & CIMA, Journal of Accountancy.
- #ai-agents
- #soc-2
- #aiuc-1
- #vendor-risk
- #ai-governance
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Tenstorrent raises $1.5B at $6.2B for open-weight inference
Tenstorrent raised $1.5B at a $6.2B valuation to build RISC-V inference servers for open-weight models. What it means for your AI hosting costs.
Read itMistral Timestral forecasts on a CPU. Your license is probably fine.
Mistral's Timestral time-series models top forecasting benchmarks, run on a CPU, and cap free use at $20M monthly revenue. Most small businesses clear it.
Read it