AI Passing the CPA Exam. Now What?
By Cody Heimerdinger
Not long ago, accounting academics ran CPA-level exam questions through early versions of generative AI and laughed at the results. The models scored poorly—around 56%—and reinforced a comfortable assumption: artificial intelligence might assist accountants, but it wouldn’t replace professional expertise anytime soon.
That assumption didn’t age well. Within a year, newer AI models were passing the CPA exam, the CMA exam, and other professional benchmarks. According to academics tracking the trend, AI’s task accuracy is now doubling roughly every seven months. Even seasoned professors have been stunned by the pace of change. The implications for accounting, tax, auditing, and financial reporting are profound.
What This Means for the Accounting Profession
The CPA exam has long served as a gatekeeper—a signal that someone possesses the technical knowledge needed to practice responsibly. AI’s rapid ascent doesn’t make the credential irrelevant, but it does change what it represents.
Passing an exam is no longer the hard part.
The real value of accountants is shifting toward:
- Professional judgment
- Critical thinking
- Skepticism
- Accountability
AI can retrieve guidance, summarize standards, and draft memos in seconds. What it cannot do is take responsibility for the outcome, or understand nuance, intent, and risk in the way regulators, auditors, and courts expect.
The Rise of “Convincing Errors”
One of the most pressing risks AI introduces is something the profession is becoming increasingly familiar with: hallucinations.
These are not obvious mistakes. AI hallucinations are polished, confident, and often sound exactly like authoritative guidance—while being subtly wrong or entirely fabricated. That makes them especially dangerous in accounting, where errors often live in footnotes, scope exceptions, and judgment calls rather than obvious miscalculations.
As a result, firms are already seeing a shift in how professionals approach research. AI is often the first stop, not the codification or traditional databases. That makes verification skills—not just research skills—mission critical.
In other words, skepticism now applies not only to clients, but to machines.
AI Is Already Writing Parts of Financial Reports
Recent academic research suggests that approximately 4.5% of new public-company disclosure text in 2024 can be attributed to AI-generated content. That may sound small, but it represents a meaningful portion of management discussion, risk disclosures, and accounting policy language.
This raises uncomfortable but necessary questions:
- Who is responsible for AI-generated disclosures?
- Should AI involvement be disclosed?
- How do auditors evaluate judgment embedded in machine-drafted language?
Many believe disclosure requirements around AI use may quickly become unnecessary—not because AI isn’t involved, but because it will be everywhere. The assumption may simply be that AI touched the process somewhere along the way.
The Promise (and Risk) of AI in GAAP and Tax Research
One of the most compelling ideas emerging from standard-setting discussions is the concept of an AI “overlay” on the FASB Codification, the Internal Revenue Code, Treasury Regulations, Revenue Rulings, and other authoritative guidance.
In theory, such a tool could democratize access to complex technical guidance—especially for smaller privately held companies without national-office resources—by quickly surfacing relevant authority, context, and connections across sources.
Done well, AI could:
- Cross-reference historical guidance
- Surface relevant interpretations
- Reduce time spent navigating dense literature
Done poorly, it could confidently invent rules that never existed. That tension defines the moment the profession is in.
Where This Is Headed (An Optimistic View)
Despite the risks, the long-term outlook is not bleak—it’s clarifying.
AI will likely handle more of the mechanics of accounting:
- Research
- Drafting
- Data analysis
Accountants will increasingly own the judgment:
- Determining what matters
- Evaluating risk
- Defending positions
- Ensuring integrity
AI may pass the CPA exam. But it still can’t sign the return, issue the opinion, or sit across the table from a regulator. That responsibility—and the judgment behind it—remains very human.