Accountants and Auditors: What the Exposure Data and the Jobs Outlook Say, and What the Big Four Report
Accountants and auditors rank 49th of 756 US occupations in Anthropic's observed-AI-exposure data, while the Labor Department data on O*NET projects faster-than-average growth to 2035. Here are both, side by side, with what EY, Deloitte and KPMG have said about AI in audit and what the UK regulator expects.
Role scorecard
Accountants and Auditors SOC 13-2011.00
- Core tasks AI does in production0 of 3
- Most mature evidenceIn production
- CostNot disclosed; EY cites a "multibillion-dollar commitment" to audit quality, technology and people (EY-stated, not AI-only)
- BenefitNo measured outcome published by EY, Deloitte or KPMG in the sources we opened
As of · Tasks: O*NET 31.01
| Core task (O*NET) | What AI or robots do | Extent | Evidence grade | Example case | Source |
|---|---|---|---|---|---|
| Prepare, examine, or analyze accounting records, financial statements, or other financial reports to assess accuracy, completeness, and conformance to reporting and procedural standards. | Agents run preliminary procedures (data extraction, evidence analysis, drafting documentation, preliminary conclusions) for professionals to review | Assists | In production | Deloitte Omnia agentic network, launched June 24, 2026 for nearly 85,000 audit and assurance professionals (Deloitte-stated) | 7 |
| Prepare detailed reports on audit findings. | Agents summarize audit documentation, such as all conclusions documented on one matter, and draft review notes; people own the decisions. This is the closest O*NET task, not an exact match | Assists | In production | EY multi-agent framework in EY Canvas, rolled out April 7, 2026 across 160,000 engagements (EY-stated, after "extensive and successful testing and piloting") | 5, 6 |
| Review accounts for discrepancies and reconcile differences. | EY expects its platform within the year to reconcile documents to external evidence and help draft standard workpapers; not yet reported as running | Not yet | Vendor claim | EY's stated plan, April 2026, as reported by Accounting Today | 6 |
Extent and grade follow our evidence grades. Both production rows are "Assists" because each firm says professionals review the output. Firms announcing tools for their own staff are both the maker and the operator, so these are operator statements, not independent measurements. Our count is 0 of 3 because no source shows an accounting task completed by AI with nobody reviewing it. We show three of the 30 core tasks O*NET lists for the occupation; the rest (tax returns, invoices, budgets and similar) have no AI case in the sources we opened.1
Costs and benefits
| Cost | No price or spend per tool is disclosed; EY describes its roll-out as part of a "multibillion-dollar commitment to audit quality, technology and people" (EY-stated)5 |
|---|---|
| Outcomes | None published: no hours saved, error rates or fee changes appear in EY's, Deloitte's or KPMG's announcements we opened578 |
| Testing | KPMG's aIQ Capture agent platform passed more than 900 technical tests, including hallucinations and prompt injection, for AIUC-1 certification (KPMG and AIUC, August 27, 2026)8 |
As of · Exposure: Anthropic, March 5, 2026; jobs data: O*NET (from BLS), 2025 and 2025–2035234
| Occupation | O*NET-SOC | Observed exposure (%) | Rank of 756 | US employment, 2025 | Projected growth, 2025–2035 | Median wage, 2025 ($) |
|---|---|---|---|---|---|---|
| Financial and Investment Analysts | 13-2051.00 | 57.2% | 9 | 377,200 | Much faster than average (7% or higher) | $102,740 |
| Accountants and Auditors | 13-2011.00 | 34.8% | 49 | 1,595,200 | Faster than average (5% to 6%) | $83,680 |
| Bookkeeping, Accounting, and Auditing Clerks | 43-3031.00 | 31.0% | 62 | 1,532,400 | Decline (-1% or lower) | $50,670 |
| Management Analysts | 13-1111.00 | 24.4% | 94 | 1,077,100 | Much faster than average (7% or higher) | $101,860 |
| Tax Preparers | 13-2082.00 | 12.8% | 161 | 95,800 | Faster than average (5% to 6%) | $54,920 |
Exposure is what Anthropic measures, not displacement: its score combines what language models could do with how they are used, weighting automated and work-related uses more heavily. Projections come from the Bureau of Labor Statistics, as shown on O*NET, and are a separate method with their own assumptions. The two sources are shown side by side, not averaged, and neither one forecasts jobs lost to AI. Rows are in order of exposure for readability; this is not a ranking of risk.34
The analysis
The role, and the jobs numbers
Accountants and auditors prepare and examine accounting records and financial statements, reconcile accounts, prepare tax filings and report to management, among 30 core tasks O*NET lists for SOC 13-2011.00.1 O*NET, using Bureau of Labor Statistics data, counts 1,595,200 of them in the US in 2025, at a median wage of $40.23 an hour or $83,680 a year, with employment projected to grow faster than average (5% to 6%) from 2025 to 2035 and 115,300 projected job openings.1
What the exposure data says
Anthropic’s March 2026 study introduces a measure it calls observed exposure, combining theoretical language-model capability with real usage data, and says AI is far from its theoretical capability.3 In its published occupation file accountants and auditors score 34.8%, 49th of 756 occupations; bookkeeping, accounting and auditing clerks score 31.0% (62nd), tax preparers 12.8% (161st) and financial and investment analysts 57.2% (9th). 411 of the 756 occupations score zero.4 The pattern across these roles is mixed: bookkeeping clerks are projected to decline while financial analysts, with the highest exposure of the five, are projected to grow much faster than average. So neither source on its own tells you what happens to a role. Anthropic itself reports no systematic rise in unemployment among highly exposed workers since late 2022, and “suggestive evidence” that hiring of younger workers has slowed in exposed occupations.3 Anthropic also notes that occupations with higher observed exposure are projected by the BLS to grow less through 2034, an average across occupations that the five finance roles above do not follow one for one.3
What the Big Four report
The audit firms describe tools that assist professionals. EY announced on April 7, 2026 a global roll-out of a multi-agent framework in EY Canvas, covering the daily workflows of 130,000 assurance professionals across 160,000 audit engagements, which it says followed “extensive and successful testing and piloting” and is expected to support all end-to-end audit activities by 2028.5 In an interview, EY’s global assurance transformation leader said agents handle tasks such as assigning work, requesting information and drafting review notes, and that auditors will still own decisions, review outputs and challenge anomalies.6 Deloitte announced on June 24, 2026 an agentic intelligence network in Omnia for nearly 85,000 audit and assurance professionals, whose agents execute preliminary procedures “for professionals to review”.7 KPMG announced on August 27, 2026 that its aIQ Capture platform earned AIUC-1 certification after more than 900 technical tests.8 All of this is the firms’ own description of their own tools; none reports a measured result.
What the regulator expects
The UK Financial Reporting Council published guidance on generative and agentic AI in audit on March 30, 2026, which it calls the first from any audit regulator. It says that “the human auditor is always accountable” and that the guidance does not change that, and its examples include summarising board minutes and using an AI tool to review contracts for revenue-recognition testing.9 It is guidance on good practice and not a response to identified quality deficiencies.9 That is the UK; we did not find an equivalent US document to cite.
What is missing
Three gaps stand out. No firm publishes how many hours, or what share of an engagement, its agents take over. None of the sources reports headcount effects: exposure is not job loss, and an announcement of tools is not a staffing statement. And all the deployment evidence is from large audit firms, which employ a small part of an occupation with 1.6 million jobs; we found no operator account for in-house accountants. For how the grades work, see how we measure AI in roles, and for the measure used here across all occupations, see occupations by observed AI exposure.
Sources
- O*NET OnLine, 13-2011.00 – Accountants and Auditors (O*NET 31.0 tasks; employment, wage and projections from the Bureau of Labor Statistics), retrieved 2026-10-28.
- O*NET OnLine, Summary reports for 43-3031.00 Bookkeeping, Accounting, and Auditing Clerks, 13-2082.00 Tax Preparers, 13-2051.00 Financial and Investment Analysts and 13-1111.00 Management Analysts, retrieved 2026-10-28 (same URL pattern with each code).
- Anthropic, Labor market impacts of AI: A new measure and early evidence, March 5, 2026.
- Anthropic (Hugging Face dataset), Economic Index, labor_market_impacts/job_exposure.csv (756 occupations), retrieved 2026-10-28.
- EY, EY launches enterprise-scale agentic AI to redefine the audit experience for the AI era, press release, April 2026.
- Gaetano C., Accounting Today, EY embeds AI agents into audits, April 8, 2026 (interview with EY’s global assurance transformation leader).
- Deloitte, Deloitte Unveils Connected Agentic Intelligence in Deloitte Omnia to Advance Audit & Assurance, press release, June 24, 2026.
- KPMG LLP, KPMG LLP Becomes First Big Four Firm with AIUC-1 Certified AI Capability, press release, August 27, 2026.
- Financial Reporting Council (UK), Innovative new guidance supports audit firm adoption of emerging AI technologies, March 30, 2026.