KPMG Draws the Line: AI for Tax Work, Not Audits
The audit exclusion in the Anthropic deal is not a technology limitation. It is a statement about where accountability still requires a human signature.
KPMG has drawn a line through its own business. On one side: tax, legal, private equity work, where Claude will handle the inference. On the other: audit, where the answer is no.
That line is the real story in KPMG's $39.8 billion global alliance with Anthropic, announced May 19. The deal covers all 276,000-plus KPMG staff worldwide and runs through the firm's Digital Gateway platform on Microsoft Azure. The initial deployment targets tax and legal client work; advisory follows by end of September 2026. Building a tax regulation AI agent that used to take weeks now takes minutes, according to KPMG.
Auditors will not see Claude integrated into their core workflows or client delivery systems — a fact KPMG confirmed to the Wall Street Journal without elaboration.
The explanation isn't complicated. Audit work sits inside some of the most consequential regulatory frameworks in professional services: PCAOB standards, auditor independence rules, Securities Act liability. An audit opinion that turns out to be wrong does not generate a client complaint. It generates an SEC enforcement action. KPMG is showing you exactly where it does not trust AI with consequential decisions — and that boundary is more informative than any capability benchmark in the press release.
The timing makes the signal sharper. Three weeks before the Anthropic announcement, KPMG cut roughly 400 US advisory staff — 4 percent of that workforce — and approximately 100 US audit partners, or 10 percent of that cohort. It also lost its largest federal audit client, the US Army, a contract held for nearly a decade at $60 million per year. The advisory roles being reduced are precisely the kind of work that now runs on Digital Gateway in minutes. The audit partners being cut are the ones who still need human judgment on regulatory attestation. KPMG is restructuring the human side of the business in the same direction as the AI deployment.
The tax and legal division being armored with Claude first is not an accident. It grew 8 percent to $9.3 billion in revenue in 2025, making it KPMG's fastest-growing and most profitable unit. Embedding AI into the highest-margin work first is a profit-center strategy. It is also a concentrated bet: if the productivity gains from minutes-per-agent compound faster than the liability exposure from AI-generated advice, KPMG extends its lead in exactly the market segment where it is winning.
The deal also gives KPMG a commercial role beyond being its own first customer. Anthropic named KPMG US a preferred consultant for deploying Claude into private equity firms and their portfolio companies. KPMG is not just using Claude. It is selling Claude, wrapped in KPMG's brand and regulatory credibility.
"Anthropic is too small to really handhold all of the companies that need the kind of very serious expertise that comes with someone that they have trusted for a long time like KPMG," Anthropic Head of Americas Kate Jensen told the Journal. Anthropic provides the reasoning. KPMG provides the client relationship and the accountability structure that regulated industries require.
That division of labor leaves an open question the press release does not answer: who bears liability when a Claude-generated tax recommendation is wrong? KPMG's announcement contains no indemnification language and no explicit AI-generated advice disclaimer. The client contracts governing that split have not been disclosed.
The weeks-to-minutes productivity claim has not been independently verified. KPMG has had Claude running internally for two years, including in its US Advisory AI and Data Labs and internal enterprise support teams. The only named external client with measurable outcomes is Keysight Technologies. No private equity firm has publicly confirmed using Digital Gateway.
Deloitte and PwC have existing Anthropic partnerships, but neither has embedded Claude into a core client delivery platform at this scale. KPMG aims to have the full deployment running by end of September 2026.
What KPMG has built is an argument about where AI can be trusted to deliver professional judgment at scale — and where it cannot. The audit exclusion is not a limitation of the technology. It is a statement about where KPMG believes accountability still requires a human signature. Everything else in the announcement is a bet that the line will move.