The Nexairi Dispatch · Friday, August 14, 2026 · Issue #35
3 Big Firms Pulled AI Reports, Then Asked for No Rules
Grant Thornton buys CBIZ for $5B, an API flaw hits three AI labs, and this week's adoption data.
Good morning, friends. The PCAOB just closed its first public comment period on AI audit standards. Three of the Big Four asked for guidance instead of rules, the same firms that have already pulled reports over AI errors. Grant Thornton answered a different question this week with a $5 billion deal to buy CBIZ. A new industry survey found nearly twice as many professionals using generative AI as last year, though almost nobody can say if it's paying off. Read on for what to check before your own auditor asks.
⚖️ AUDIT STANDARDS — Big 4 Ask PCAOB for Guidance, Not Rules
What happened: The PCAOB closed its first-ever public comment period on AI audit standards August 7 with 33 letters filed. PwC, EY, Deloitte, KPMG, Grant Thornton and the AICPA all asked for guidance rather than formal rules.
Why it matters: Three of those same firms, KPMG, Deloitte and EY, have each already pulled or refunded a published report after AI hallucinations turned up inside it. Asking for flexibility while cleaning up your own AI mistakes is a specific position to take. It tells you the industry still doesn't agree on what "checked" means.
What to watch: Formal PCAOB guidance is still coming. Firms that build their own AI documentation now will have less to redo when it lands.
Outside Nexairi
Grant Thornton Buys CBIZ for $5 Billion — Journal of Accountancy
The all-cash deal creates the fifth-largest professional services firm in the U.S. It lands right in the middle of the AI pressure squeezing the traditional compliance business model.
AI Won't Kill Accounting, But It May Kill the Business Model — CPA Practice Advisor
AI now produces the routine compliance work firms used to bill for, fast and cheap. It's the production, not the professionals, that's disappearing as a paid service.
A Shared Flaw Let Weak AI Models Read Strong Ones' Reasoning — The Hacker News
An API flaw touching OpenAI, Anthropic and Google let weaker models decode how stronger models were reasoning through a problem before it was patched. It's a reminder that vendor boundaries aren't guaranteed just because a vendor says so.
Anthropic Posts First Profit on $10.9 Billion in Revenue — AIToolsRecap
Anthropic turned its first quarterly profit as OpenAI's own IPO paperwork is reportedly close behind. Both companies are shifting from growth-at-any-cost toward proving the business actually works.
Firms Say They Want AI. Few Can Actually Run It. — CPA Practice Advisor
A new survey finds accounting teams split sharply between heavy AI users and firms with no real plan. Most can't say whether their AI spending is paying off.
Tool Worth Knowing: Puzzle (puzzle.io)
Puzzle is AI-native accounting software built for startups and the firms that serve them, with a direct Stripe API connection that automates revenue recognition instead of matching transactions by hand. Worth a look if you have SaaS clients drowning in manual Stripe reconciliation.
Deeper Read
2026 AI in Professional Services Report: AI Adoption Has Hit Critical Mass, But Now Comes the Tough Business Questions — Thomson Reuters Institute
The full report behind this week's adoption numbers, including the ROI measurement gap the headline stat leaves out.
Firms, Accountants Divided on AI But Perhaps Not for Long — Accounting Today
A look at the split between firms leaning hard into AI and those still waiting, and why that gap probably won't hold much longer.
Quick Hits
- OpenAI names Dali Rajic Chief Revenue Officer, its second in under a year
- OpenAI expands ChatGPT ad tests to the UK, Mexico, Brazil, Japan and South Korea
- Anthropic names former California Supreme Court justice Tino Cuéllar its first Chief Global Affairs Officer
- Intuit launches its AI-native Accountant Suite with cross-client insights and custom automations
- Forrester finds enterprises are postponing 25% of planned AI spend to 2027
- Puzzle adds Stripe-native revenue recognition to its AI bookkeeping platform
- Dext claims 99.9% accuracy on receipt scanning. Our review of real Capterra ratings tells a fuller story
- AI absorbed the fractional CFO's reporting work. The advisory work is what's left to re-price