Generative AI is part of how college students today study and complete their work. In Inside Higher Ed’s 2025 Student Voice survey of 1,047 students, 85% said they had used AI for coursework in the past year. A critical challenge in the age of AI is that accounting faculty and the employers who hire their graduates need to know whether students are still learning the material and whether those students can tell when an AI tool is wrong.
Hussein Issa of Rutgers Business School addressed those questions in a September 2026 presentation for the AICPA’s Government Performance and Accountability Committee (GPAC). Issa is an associate professor of Accounting & Information Systems at Rutgers Business School and a faculty member of its Continuous Auditing & Reporting Lab (CAR Lab). He leads the school’s work to build AI into its curriculum and received the American Accounting Association/Deloitte Foundation’s 2026 Cook Prize for excellence in teaching accounting. His talk focused on teaching, but much of it applies for anyone training new staff.
What AI can do for faculty and students
Faculty can use generative AI to keep course materials current, draft cases and exam questions, and build realistic simulations and open-ended assignments, while the professor in the classroom remains ultimately responsible for the content delivered to students. AI tools can produce a first draft, but the faculty member should decide what to use, including any feedback that goes to students.
For students, AI use cases include tutoring on hard topics, getting past a blank page, feedback on writing, help with software and code, and practice for job interviews. Several of these use cases focus on access; for instance, a student working long hours or still learning English can get help at midnight, long after office hours end.
The risks of AI go beyond cheating.
Cheating is a serious risk and potential issue with AI use. For the profession, however, two other issues also stand out as posing serious risk.
The first is deskilling, meaning people never build skills, or slowly lose them, because a tool does the work. A 2025 study in the Proceedings of the National Academy of Sciences followed nearly 1,000 high school math students. Students given an unrestricted GPT-4 chatbot did 48% better on practice problems. Once the tool was removed, they scored 17% worse on exams than students who never used it.
The students using AI were not really learning. A version of the AI tool that gave hints instead of answers largely avoided that drop in exam performance. In accounting terms, a student who lets AI do the bank reconciliation can turn in the right answer and still not know how to do one, which can be problematic when a circumstance arises that requires professional judgment that is built over many hours of hard work.
The second problem for the profession is what Issa called “a market for lemons,” a theory based on used cars from economist George Akerlof . When buyers can’t tell good cars from bad ones, every car sells at an average price, and owners of good cars lose out. The same thing can happen with grades. If a polished memo or a strong GPA no longer tells an employer what a graduate can do, the best students lose the most and employers end up making worse hiring decisions. Employers depend on those signals, so they have a stake in keeping them reliable.
Issa also flagged accuracy and privacy, including the risk of confidential data leaking out through AI tools. A student who gets used to pasting class data into a public chatbot may do the same with client data on the job. It is much better to learn in school that confidential information never goes into a public AI tool than to learn after a breach on the job.
Assistant or actor?
Most AI tools in use today are assistants. They answer a question, and a person decides what to do with the answer. However, more recent “agentic” AI tools go further. They can carry out a series of steps on their own, such as pulling data, matching transactions, drafting entries, and sending them for approval. Issa’s current research looks at how organizations should oversee AI agents that act less like tools and more like participants in a process.
For accountants, basic control questions get harder when software does the work. Who authorized the transaction? Who reviewed it? Is there an audit trail? An approval from someone who never grappled with hard questions and truly internalized the work isn’t much of a control.
A 2025 Government Accountability Office (GAO) review found that 11 federal agencies reported 282 generative AI use cases in 2024, up from 32 in 2023. GAO’s AI Accountability Framework draws on federal internal control standards and includes audit procedures for each key practice. That is familiar work for public sector CPAs, so they have good reason to help hold government accountable for how these systems are used. However, as AI takes on more of the work, students and new staff need to learn to check what it produces and take responsibility for the result.
What educators can do
Educators should have students check AI’s work. For example, give students an AI-drafted journal entry, reconciliation, or memo with planted errors, and grade them on what they find and how they explain it.
An April 2026 session of the AICPA & CIMA Faculty Hour webinar series shared activities like this, including exercises where students correct errors in AI-generated journal entries. These activities can be adapted and work for large classes.
Educators can also test what students can do on their own. Have them work problems in class and pick a few at random to walk you through their written work. That shows what they understand in a way that a finished, potentially AI-generated, product can’t.
What practitioners can do
Practitioner input in the classroom is also critical now, as experienced professionals, faculty, and students alike navigate the transition to AI. For example, practitioners can inform faculty and students what entry-level work looks like now. Good places to do that include guest lectures in accounting classes at universities and colleges and involvement with university advisory boards.
Public sector practitioners have an advantage because budgets, annual comprehensive financial reports, and audit reports are public. Students can use AI to analyze a city’s finances and then defend their conclusions to a public sector CPA.
Additionally, practitioners should rethink how new staff learn. If AI takes over the tasks that used to teach new hires the basics, firms and agencies will need to teach those basics in other ways. For example, consider developing simulations or having new staff do tasks (such as reconciliations) themselves before reviewing AI-prepared ones, then ask them to walk you through what they find.
Preparing future CPAs to supervise AI
In accounting, AI will increasingly be used both as an assistant and an actor. But students still need to learn the work themselves and cannot depend entirely on AI. The goal should be to train CPAs who know accounting well enough to catch an AI tool’s mistakes and understand that they still take responsibility for the work.