Finance & Accounting AI Briefing — Stone Arch Collective

The Stone Arch Bridge spanning the Mississippi River in Minneapolis — Stone Arch Bridge, Minneapolis — via Wikimedia Commons (placeholder — swap for one of ours).

Today's Ledger Line

With the Oct. 15 extension deadline bearing down, CPA Practice Advisor's new reporting on firm AI tools hitting usage ('token') caps mid-crunch — pushing stressed staff toward personal, unapproved chatbots loaded with client financial data — turns 'shadow AI' from an IT nuisance into a live client-confidentiality and PCAOB-documentation problem that firms need an overflow policy for before the deadline, not after.

Today

Finance & Accounting AI News

01
A CPA Practice Advisor piece describes the scenario directly: it's early October, a senior associate is racing the Oct. 15 extension deadline using the firm's AI assistant to summarize client documents and draft emails, hits a usage limit with three returns still waiting, and opens a free AI chatbot on a personal laptop to keep going with a client's brokerage statement. The piece frames this as a token-limit-driven 'shadow AI' exposure firms haven't priced into their risk policies. CPA Practice Advisor
02
Accounting Seed launched what it calls a 'Reimagined Experience' built around AI; the AI model builds knowledge of each business's workflows, approval paths, policies, and revenue recognition rules, then applies that context across accounts-payable and close workflows rather than operating as a bolt-on chat tool. CPA Practice Advisor
03
Deloitte debuted an AI-enabled M&A platform aimed at tax and transaction teams; capabilities help customers map products to tax codes, report transaction and tax liability data, validate exemption certificates, auto-generate tax matrices, analyze results across multiple applications, and configure custom rules, with availability after October 8. Accounting Today
04
Accounting-AI startup Accrual launched its agent platform, Arc, with $75 million in funding; co-founder Siddarth Chandrasekaran said the breakthrough isn't simply giving accountants access to a more capable AI model — it's giving that model the context, tools, controls, and accounting-specific understanding required to do real work. Firms can teach Arc processes like reviewing tax notices against client records and firm templates, then save the workflow as a reusable agent other staff can run or schedule. CPA Practice Advisor

Past 7 days

September 30, 2026

01
AuditDashboard, a client engagement management platform, rolled out AI Connectors that give accounting firms the ability to connect Claude, ChatGPT, or Microsoft Copilot directly to their engagements and live PBC data — putting external large language models (LLMs, AI systems trained on text to generate human-like responses) directly into the client-file pipeline auditors are already being asked to document. CPA Practice Advisor
02
Canopy unveiled an external API platform designed for accounting firms that want to integrate, automate, and build on Canopy across their tech stack, securely sharing core data and workflows with other internal systems, another step toward ambient AI sitting inside everyday practice-management tools rather than as a bolt-on. CPA Practice Advisor
03
Accounting Seed touted a reimagined, AI-built experience in which its AI model builds knowledge of each business's workflows, approval paths, policies, and revenue recognition rules, then applies that context automatically — a sign small and mid-size firms will increasingly inherit AI configuration decisions baked in by their general ledger vendor rather than choosing them separately. CPA Practice Advisor
04
MindBridge expanded its finance and audit platform this week with Agentic Risk Assessment and Financial Close Oversight; CEO Les Rechan said "Customers are telling us they want AI to reduce complexity, not create more of it" and that the new tools are built to keep people in control of decisions they're accountable for. The Agentic Risk Assessment capability connects evidence, analytics, assertions, materiality, and methodology to help auditors make faster, better supported, and more defensible audit planning decisions. CPA Practice Advisor
05
Politico reported that AI agents are becoming a genuine worry for bank deposit stability, with financial firms, regulators and lawmakers still grappling with how the technology might affect depositor behavior; Rep. Bill Foster (D-IL) told the outlet "There's a whole class of worries there that I've been worried about for a while." The piece signals systemic-risk scrutiny is starting to catch up to agentic AI in finance, a dynamic that will eventually shape what controls regulators expect around client-facing finance agents too. Politico (via LLRX's AI in Finance and Banking roundup)

September 30, 2026

01
IBM's CFO study, conducted with Oxford Economics across 1,500 CFOs in 33 geographies, found the CFO role is expanding well beyond the balance sheet, with more than half of respondents reporting greater portfolio-management authority and capital-reallocation responsibility tied to AI strategy — but IBM's CFO Jim Kavanaugh warns most companies fail because they deploy AI tools before fixing data architecture and redesigning workflows. Fortune
02
A new report from finance-software maker insightsoftware found that concerns about AI-driven job displacement in finance are climbing, with 20% of finance professionals now fearing AI will replace their jobs, up 7 percentage points from last year — pressuring CFOs to communicate more clearly with staff about how AI will change roles. CFO Dive
03
U.S. Bank's CFO Insights Report: Fall 2026, surveying 1,000 senior finance leaders, found CFOs growing more optimistic about their businesses' financial prospects even as many express anxiety over unpredictable AI spending and budget overruns; cost-cutting and efficiency remain the top stated priority at 37%. CFO Dive / Accounting Today
04
Visa has open-sourced part of its AI-powered cyber-defense system after what it called "humbling" lessons from vulnerabilities exposed by AI models, as the payments giant — which processes roughly a billion payments a day worth around $15 trillion a year — prepares for autonomous, AI-driven cyberattacks. Thomson Reuters (via LLRX)
05
A new industry analysis on multi-agent systems in banking argues the pressing question is no longer whether AI agents can act on financial systems but who is accountable when a chain of specialized agents — one retrieving data, one analyzing, one drafting, one executing — makes a costly error. LLRX (AI in Finance and Banking roundup)

September 28, 2026

01
AI-powered talent marketplace Mavi officially emerged from stealth today, with co-CEO Molly Liu telling TechCrunch that the U.S. accounting profession is shrinking and that many remaining practitioners aren't fluent in the AI skills firms now need, while the service connects and onboards U.S. companies with AI-proficient global financial and accounting talent within days. TechCrunch
02
Auditoria.AI's seventh annual State of AI Automation in the Finance Office report found that 66.5% of finance organizations are increasing their AI investment, with 24.2% now treating it as a top budget priority, while invoice/document digitization, anomaly detection, cash-flow forecasting, and compliance reporting are common use cases with no single one dominating adoption — evidence that spend is rising faster than any consensus on where to point it. GlobeNewswire / Auditoria.AI
03
Speaking at Workiva's Amplify conference, CFO Barbara Larson said finance is now being pulled into strategic decisions earlier than ever, with panelists framing the real shift as finance teams having to build the trust and guardrails that let a company move fast on AI without breaking something; as one exec put it, "Speed is not an advantage if the trust isn't there." Thomson Reuters Checkpoint
04
A new practitioner guide confirms the SEC has not issued a standalone AI disclosure rule as of September 2026, but its Division of Corporation Finance has sent at least 92 AI-related comment letters to 56 companies between 2021 and October 2024, with volume accelerating through 2025 and 2026 — meaning existing anti-fraud statutes, not new rules, remain the operative risk for any AI capability claim in a filing. Finrep.ai

September 25, 2026

01
Gartner's latest AI-in-finance research, unveiled Thursday, found 55% of CFOs reported positive overall returns from their 2025 AI initiatives, but when asked about individual AI use cases, 57% said returns were unclear, and analysts said finance teams must set realistic time-to-value expectations for their artificial intelligence investments as pressure mounts to deploy the technology more strategically. CFO Dive (via Yahoo Finance)
02
Feedzai, the global leader in AI-native financial crime prevention, launched Farol, an embedded AI agent engineered to slash fraud investigation latency and sharpen threat detection across risk operations in banking, arriving as fraud grows costlier to fight on every front, with mandatory reimbursement rules increasing loss exposure while technology spend and manual review costs keep climbing. Business News This Week (PRNewswire)
03
Emburse this week announced the launch of Emburse AP, a new AI-powered accounts payable and payments solution for growing businesses and organizations, which brings invoice processing, approval workflows, vendor enablement and payments together in one intuitive platform to give finance teams greater visibility and control over every dollar spent. Dwealth.news (AI & Finance weekly roundup)
04
TechCrunch profiled Tabby, a startup from a former accountant who was a successful accountant, overseeing a team of 20 people and handling more than 2,000 returns a year before building AI tooling to fix accounting software that was rarely designed with actual accountants in mind — an early sign that practitioner-built AI is starting to displace legacy practice-management software rather than just automate tasks inside it. TechCrunch
05
The AICPA's September Journal of Accountancy warns that AI can speed tax research, but it can also invent cases, so firms need guardrails to avoid potential trouble, alongside coverage noting a new generation of agentic AI tools can plan, execute, and adapt as they complete tasks, raising expectations for what automation can do inside firms and finance departments. Journal of Accountancy (AICPA & CIMA)

September 24, 2026

01
The AICPA and Danny Werfel formally stood up the Council on AI Risk in Tax (CART) on September 22-23, with participants from accounting, law, technology, government, and academia, including the National Association of Enrolled Agents and the Federation of Tax Administrators, tasked with stress-testing Werfel's AI Risk Framework for Tax. Werfel put the stakes bluntly: "Tax is high stakes, and we cannot assume that emerging AI tools will police themselves. The AI risk framework published earlier this month provides a starting point." Journal of Accountancy
02
A new Deloitte/Center for Audit Quality survey released this week found the audit committee's share of primary responsibility for AI governance grew from 20% to 29% in the latest annual survey by Deloitte and the CAQ, even as audit committees are increasingly looking at growing risks such as artificial intelligence without necessarily having the technical expertise to evaluate it. Accounting Today
03
Accounting Today's governance column flagged that 81% of tax and audit professionals are now utilizing AI tools at least several times a week, but surprisingly, 35% say they use tools that their firm hasn't authorized for work, a gap the piece calls informally called "shadow AI" that leaves engagements with no log of what the tool was given or whether anyone checked its output. Accounting Today
04
Former PCAOB member Christina Ho, who resigned from the audit oversight board in January 2026 after a career of dissenting on new audit rules, is now hoping to reshape the audit profession in a different way: by joining a startup, Oath Verified, that aims to use AI to make audits more accurate, effective, and, yes, cheaper — and it's structured as a licensed CPA firm, not just a SaaS vendor. Going Concern

September 21, 2026

01
At Workiva's Amplify conference on September 14, Workiva CFO Barbara Larson said finance leaders are now being pulled into strategic decisions before they're made rather than after, and are increasingly asked to lead business, tech, and data strategy — while Northwest Bankshares CFO Doug Schosser argued data ownership has to sit with whoever creates the data, not be treated as an IT problem. Thomson Reuters Tax & Accounting
02
The Journal of Accountancy's September 2026 issue warns that AI can speed up tax research but can also invent cases outright, and tells firms they need documented guardrails to avoid the kind of trouble that's already produced a string of court sanctions over fabricated AI citations. Journal of Accountancy
03
Accounting Today's September 4 tech roundup reported that Deloitte launched a new practice built around open-source AI models and that governance-risk-compliance (GRC) platform LogicGate shipped its 2026 release adding AI "agents" — semi-autonomous programs that plan and execute multi-step tasks — for compliance and lifecycle-management workflows. Accounting Today

Regional

Upper Midwest Watch

01
Members Cooperative Credit Union, a $1.2 billion institution based in Duluth, Minnesota, stands out in national AI-adoption reporting for its internal AI usage policy; Members Cooperative Credit Union ($1.2B, Duluth, MN) has a robust internal policy that provides guardrails for usage — an example of a Minnesota credit union moving past ad hoc AI experimentation toward written employee guardrails. CUToday / creditunions.com
02
In Wisconsin, Marine Credit Union implemented AI-powered auto-decisioning that now handles 55-60% of consumer loan applications with only a 3% variance from human loan officer decisions, and the credit union has since lent at a higher rate than its peer group. Evok Credit Union Marketing

Past 7 days

September 24, 2026

01
The Wisconsin Institute of CPAs' Business & Industry Fall Conference this week promised attendees they'd "hear from leading experts on the latest tax developments, AI and cybersecurity, business transactions and valuation, workforce talent strategies", aiming it squarely at CFOs, controllers, and accounting staff across the state. Wisconsin Institute of Certified Public Accountants (WICPA)
02
The Iowa Society of CPAs has opened registration for its AI for CPAs Series, offered as a bundled webinar package with multiple sessions between Friday, October 16 and Wednesday, December 30, 2026, giving Iowa practitioners a structured, CPE-bearing on-ramp to AI governance rather than ad hoc tool adoption. CPA Practice Advisor (State CPA Society News & Updates)

Practice

Try This on the Books

Stress-Test Your Firm's AI Overflow Plan

The 10-minute versionAsk two or three staff who've been using the firm's licensed AI assistant this week whether they've ever hit a usage or token cap, and what they did next. Listen specifically for any mention of switching to a personal or free AI account to keep working with client data.

Level upDraft a one-paragraph addendum to the firm's AI policy that pre-approves a second, client-data-safe AI tool for overflow capacity during extension season, and explicitly bans personal AI accounts for any work involving client financial documents.

Learn more — Read the shadow-AI token-limit story

Build Your AS 1105 Calibration One-Pager

The 10-minute versionFor any engagement where the team used AI-assisted anomaly detection in journal-entry testing, write three bullets: what data trained or calibrated the tool, how false positives were evaluated, and who on the engagement team signed off on the output.

Level upTurn the one-pager into a reusable workpaper template and add it to the firm's audit methodology manual so every engagement using AI-assisted testing documents the same three items before the next PCAOB inspection cycle.

Learn more — Review the PCAOB technology-assisted analysis standard

Past 7 days

September 30, 2026

Write the three-line AI audit trail before an inspector asks for it

The 10-minute versionPick one anomaly your firm's AI-assisted journal-entry testing flagged this month. In three lines, note: what data trained or calibrated the model, how false positives were evaluated, and who signed off on the result — the exact elements PCAOB inspections have started checking regardless of whether the AI's output was correct.

Level upTurn the three-line note into a standing template attached to every engagement file that uses AI-assisted analytics, so the documentation exists by default rather than being reconstructed after an inspection request.

Learn more — PCAOB's existing standards on AI-assisted audits

Give one AI tool a two-line, tool-specific ROI test

The 10-minute versionFor one AI tool your firm already uses or is evaluating, write down: the single task it automates, the hours saved per week, and the dollar value of that time. Skip the feature list — CFOs are now asking for tool-specific proof, not portfolio-level averages.

Level upCompare your case against what MindBridge's newly launched Financial Close Oversight and Agentic Risk Assessment features promise out of the box, to see whether your tool's ROI case is keeping pace with what the market now expects as standard.

Learn more — MindBridge's new agentic finance and audit capabilities

September 30, 2026

Run IBM's Three-Prerequisite Gut Check

The 10-minute versionPick one workflow you're considering for AI (close, AP, tax research) and score it 1-5 on each of IBM's three prerequisites: data architecture/governance, workflow redesign, and technology deployment readiness — most firms fail because they jump straight to the third without the first two.

Level upShare the scorecard with whoever approves your next AI tool purchase and flag the weakest-scoring prerequisite as a precondition before the contract is signed.

Learn more — Read the IBM CFO study coverage

Document One AI-Assisted Test Like an Inspector's Watching

The 10-minute versionTake one AI-assisted anomaly-detection or journal-entry test your team ran this week and write three sentences in the workpaper: what data calibrated the tool, what it flagged, and how false positives were evaluated — the exact gap PCAOB inspectors are now citing under AS 1105.

Level upTurn those three sentences into a standing workpaper template so every future AI-assisted procedure gets the same documentation without extra thinking time.

Learn more — See PCAOB AI documentation expectations

September 28, 2026

Pressure-test one AI feature's ROI, not your whole stack

The 10-minute versionPick a single AI-assisted task you already use (transaction categorization, invoice matching, anomaly flagging). Pull your last 5 instances and log actual minutes spent versus your best estimate of the pre-AI manual time. Write the delta as one sentence: 'X saved Y minutes on Z task.'

Level upTurn that sentence into a one-page, tool-specific ROI case (time saved x hourly cost, annualized) you could hand a CFO cold — the specific ask this week's news says every vendor and internal champion now needs.

Learn more — See the CFO-trust angle behind the ROI demand

Check one AI claim against the SEC's 'AI-washing' bar

The 10-minute versionGrab one public AI capability claim — a website line, pitch deck slide, or draft filing sentence — from your firm or a client. Ask: does the actual tool do exactly what this sentence says, or is it rounding up? Note any gap in one line.

Level upBuild a simple two-column 'AI claim → supporting evidence' tracker for every AI mention headed into a 10-K, marketing piece, or engagement letter, so you have a documented answer ready if a comment letter or client ever asks.

Learn more — Read the SEC AI-disclosure practitioner guide

September 25, 2026

Stress-test one AI tool's evidence trail

The 10-minute versionPick one AI-assisted process you already use (close checklist, anomaly detection in journal-entry testing, AP matching). In 10 minutes, write down: what data trained or calibrated the model, who reviewed the last batch of flagged exceptions, and where that documentation lives. If you can't answer in under two minutes, that's your inspection-risk gap.

Level upTurn the answers into a one-page 'AI model card' template (tool name, training/calibration source, false-positive review cadence, sign-off owner) and require it for every AI tool before next quarter's close.

Learn more — See ICAEW's AI-fraud verification checklist

Run a 10-minute use-case ROI gut check

The 10-minute versionList every AI tool or pilot your firm or client currently pays for. Next to each, write the one metric it was supposed to move (hours saved, error rate, days-to-close) and whether anyone has actually measured it since go-live. Circle any tool with no measurement — that's the one a CFO or partner will ask about first.

Level upFor the top circled tool, draft a 90-day measurement plan: baseline metric, target, and who owns tracking it, then bring it to next month's budget conversation instead of a renewal invoice.

Learn more — Read Gartner's CFO AI ROI findings

September 24, 2026

Run a mini AS 1105 calibration check on your AI anomaly-detection output

The 10-minute versionPull last month's AI-flagged journal-entry anomalies. For the first five flags, write one sentence each documenting what triggered the flag, what rule or model logic produced it, and how you confirmed or dismissed it.

Level upTurn those five sentences into a one-page 'AI calibration memo' template with fields for training data source, false-positive threshold, and reviewer sign-off, then require it firm-wide before the next inspection cycle.

Learn more — How AI is changing audit anomaly detection

Surface your team's shadow AI

The 10-minute versionAsk three staff members which AI tools they personally used on client work this week that weren't on the firm's approved list, and log the tool name plus which engagement it touched.

Level upDraft a one-page approved-AI-tools list, circulate it for signature acknowledgment, and set a recurring 90-day refresh so new tools get vetted instead of quietly adopted.

Learn more — Read the 'shadow AI' governance piece

September 21, 2026

Run a 10-minute AI citation spot-check

The 10-minute versionPull one AI-assisted tax research answer or memo drafted this week and independently verify every case cite, code section, and regulation reference against the primary source before it goes near a client or the IRS.

Level upTurn the spot-check into a standing sign-off line on your engagement checklist — no AI-assisted research memo leaves review without an initialed 'citations verified against primary source' box, tying directly to the Circular 230 due-diligence duty.

Learn more — Journal of Accountancy: AI can speed tax research, but it can also invent cases

Document one AI-flagged anomaly the way an inspector would want

The 10-minute versionTake one transaction or journal entry your AI anomaly-detection tool flagged this month and write a short paragraph covering how the model was calibrated, what data fed it, and how the false-positive rate was assessed.

Level upBuild a one-page 'AI tool documentation' template from that paragraph and require it in every engagement file where AI-assisted analytics touch evidence, so you're not reconstructing the record after an inspector asks.

Learn more — CalCPA: AI Standard Setting, Regulation in Audit

Implications

What This Means for Practitioners & Clients

01Extension-deadline crunch plus hard token caps on firm-licensed AI tools is a predictable, recurring trigger for staff to move client data onto personal, unapproved AI accounts — firms need an approved overflow tool and a written policy in place before Oct. 15, not as a post-incident fix.
02Vendors like Accounting Seed are embedding AI directly into existing close and AP workflows rather than selling a separate chat tool, which shifts the ROI conversation practitioners need to have: it's no longer 'which AI tool should we buy' but 'which specific workflow does this replace, and what's the measurable time savings per engagement.'
03Deloitte productizing AI for M&A tax mapping and transaction analysis (available Oct. 8) signals that client expectations for AI-accelerated due diligence will rise even at firms that don't do Big Four-scale deals — mid-size firms offering M&A advisory should expect to be benchmarked against that speed.
04Upper Midwest credit unions are already past the pilot stage — Duluth's Members Cooperative has a written AI usage policy and Wisconsin's Marine Credit Union is running AI auto-decisioning on a majority of consumer loans — so accountants and auditors serving these institutions should expect AI governance and model-risk questions to show up in engagements now, not as a future hypothetical.

Watch

Audit & Assurance Watch

PCAOB Technology-Assisted Analysis Amendments (AS 1105 / AS 2301)Standard: AS 1105 / AS 2301 (Technology-Assisted Analysis)
The PCAOB adopted amendments clarifying auditors' responsibilities when using technology-assisted analysis, updating AS 1105 (Audit Evidence) and AS 2301 (Auditor's Responses to Risks of Material Misstatement) to address procedures that use computers and algorithms to analyze electronic information. Auditors remain responsible for designing and supervising procedures even when using automated tools.
PCAOB / Houseblend summary
PCAOB Policy Statement on AS 1105.10A — Reliability of External Electronic InformationEffective: FYs beginning on/after Dec. 15, 2025
On September 18, 2025, the PCAOB issued a policy statement to guide implementation of paragraph .10A of AS 1105, Audit Evidence, which takes effect for audits of fiscal years beginning on or after December 15, 2025. Firms using AI tools to pull or validate client-provided (PBC) data now fall squarely under this reliability-evaluation requirement.
Center for Audit Quality
SEC AI Task ForceStatus: Active, SEC-wide coordinating body
The SEC launched a task force on artificial intelligence to spearhead the agency's efforts to enhance innovation and efficiency through responsible AI use, centralizing cross-agency and cross-disciplinary collaboration to navigate the AI lifecycle and maintain governance. Registrants and their auditors should expect this body to shape future AI-governance disclosure expectations.
Center for Audit Quality

Weekly — updates Mondays

Regulatory & Standards Watch

Last updated: September 28, 2026

Legislation

OPR Alert 2026-19 (Circular 230 AI Guidance)

Federal — IRS Office of Professional Responsibility · In effect since June 24, 2026; contested by AICPA over billing language (ongoing since Sept. 15)

Applies six existing Circular 230 duties — due diligence, competence, confidentiality, and prohibition on unconscionable fees among them — to practitioners' use of generative AI in tax work, without creating new rules.

Read the bill

AI Disclosure Rulemaking Petition, File No. 4-882

Federal — SEC · Filed February 2026; no rulemaking action taken, considered unlikely under current SEC leadership's deregulatory posture

Would require mandatory AI governance disclosure rules for public companies rather than relying on existing anti-fraud statutes.

Read the bill

Cyber and Emerging Technologies Unit (CETU)

Federal — SEC · Active since February 2025; coordinating with CFTC under a March 2026 memorandum of understanding on joint enforcement and examinations

Dedicated enforcement unit investigating 'AI washing' — companies overstating or mischaracterizing AI capabilities in investor-facing communications.

Read the bill

Grants

Nothing here yet.

A note on accuracy: this briefing is AI-assisted research. Double-check anything you'd cite, quote, or act on — especially numbers, dates, and direct claims.