Ernst & Young tells California Board of Accountancy AI can reshape audits but raises governance and trust questions

Ernst & Young partners presented a detailed briefing to the California Board of Accountancy on Nov. 20, 2025, arguing that generative artificial intelligence is already changing how audits are done and will continue to reshape assurance work, workforce skills and audit governance.

The presentation — led by Richard Jackson, U.S. Assurance chief technology officer, and Ryan Liu, West Region Assurance AI leader — laid out common enterprise use cases (document intelligence for invoice and contract processing, full-population analytics for risk identification, automated tie‑outs of financial statements and ‘code explainer’ tools for reviewing application controls). Jackson said the profession is “in an inflection point” since generative models entered public view and called AI “a sea of contradictions,” noting high reported returns on investment alongside research that many AI projects fail.

Board members and presenters focused on three practical questions: how firms will build trust in AI outputs, what controls and training are required before auditors reduce human oversight, and how standards and fees will adapt. Jackson and Liu described firm-level controls used at Ernst & Young: internal firewalls around models, controlled deployments (for example, Copilot within the firm), supervisory review requirements and structured training and feedback loops so models can be refined (sometimes called RAG—retrieval-augmented generation—training).

Concern about cybersecurity and so‑called "agentic AI" was a recurring theme. Jackson noted a recent public case study of an automated AI cyberattack that required AI-based detection to intercept, warning that greater autonomy in AI agents expands the cyberattack surface.

Several board members pressed on audit practice implications. One member asked whether AI-driven efficiencies mean lower audit fees; presenters answered that transparency and client‑level conversations are necessary and that the economics depend on the scope of services and the level of human oversight retained. On field work, presenters said AI will shift the focus of in-person engagements toward discussion of anomalous findings and professional judgment rather than eliminating boots-on-the-ground testing.

Presenters also discussed regulation and standards. They pointed to the National Institute of Standards and Technology (NIST) as a leading voluntary framework that many U.S. organizations use to structure governance and urged the profession and standard‑setters to interpret existing auditing standards for an AI environment rather than assume an immediate need for brand‑new standards. Still, they acknowledged areas—such as moving from sample testing to population analytics—where auditing practice guidance may need updating.

Board reactions mixed cautious optimism with skepticism about the pace of change. Several board members said they are uncomfortable asking consumers to "trust us" if audits are completed faster with less visible human review; presenters urged greater transparency with audit committees and suggested disclosing the role and limits of AI in engagements.

What’s next: the board discussed folding further AI work into existing committees and continuing education planning, and several members asked staff to return with focused materials on how AI affects continuing education, enforcement and peer review oversight. The presentation closed with an offer by presenters to continue engagement and provide examples of how firms of differing sizes can adopt AI with controls in place.

The presentation ran from the introduction of the E&Y guests through an extended Q&A (presentation began at the meeting’s AI item and concluded before lunch). The board recessed for lunch after the session and scheduled follow-up discussions in future meetings.