Five stories this week, in priority order.
Story 1: AI verification still not operationally mature
EY published a survey on September 15 of 202 senior AI decision-makers at publicly traded companies with more than $1 billion in revenue, fielded May 28 to June 15, 2026. 91% said their organization uses agentic AI, in pilots or in full deployment. Of those, 85% admit that at least some of these systems execute actions without real-time human involvement. 98% have formal AI governance policies and run an assurance review at least annually. Those reviews have teeth: 25% of organizations fully stopped a quarter or more of their AI systems after one. 69% are concerned their organization lacks the internal expertise to evolve its AI governance controls, and 36% report an AI incident or failure that caused a materially negative impact.
Issue 6 covered AI-fabricated case citations reaching a Tax Court filing. The exposure here is different: the output never leaves the building. A forecast adjustment or an accrual that an agent posted without review is consumed internally, so there is no opposing counsel to flag it.
Why it matters: these are billion-dollar public companies with written policies and annual reviews, and a quarter of them still had to shut down a large share of their AI systems. A mid-market finance team with fewer controls should assume its agents act unreviewed more often.
What to do this week: ask whoever owns your finance AI tools (forecasting, close automation, AP matching) which of them approve, flag or post without a human review step, and when that step was last tested. Then ask which of those steps the agent performs by generating a number and which it performs by calling a deterministic tool, a calculation, a lookup or a rule that returns the same answer every time. Anywhere the agent generates a figure that a tool could compute, that could be the fix.
Source: EY newsroom, "Autonomous AI implementation outpaces oversight," Sept. 15, 2026 (202 respondents, fielded May 28–June 15, 2026)
Story 2: AI policy is the weakest policy area in treasury
AFP released its 2026 Treasury Benchmarking Survey Report on September 15, sponsored by PNC Bank, based on 425 treasury practitioner responses from May 2026. Respondents scored their own policies for effectiveness. Cash management scored 4.4 out of 5 and bank relationship management 4.3. AI and emerging technology policy scored 2.9, the lowest of any area measured. The skills side matches: 34% rate themselves effective in AI knowledge, against 50% who consider it important. At the same time, AI and automation entered treasury's top five priorities this cycle at 30%, and managing AI opportunities and risks (38%) and using AI to automate manual processes (35%) now rank among treasury's most significant challenges.
Why it matters: treasury teams are adding AI to the priority list faster than they are writing the policy that governs it. Story 1 shows where that leads.
What to do this week: pull up your written AI and automation policy for treasury systems. If there isn't one, that is the gap, ahead of any tool evaluation.
Source: AFP, 2026 Treasury Benchmarking Survey Report, released Sept. 15, 2026 (PNC Bank-sponsored; 425 respondents, fielded May 2026)
Story 3: Contractual obligation tracking appears as potential AI agent use case
Agiloft released a study on September 15 of 2,005 senior procurement leaders at U.S. and U.K. enterprises, fielded August 5 to 20, 2026. 97% are confident they systematically capture the full value of what they negotiated. In the same survey, 52% failed to collect a contractual remedy in the past year because it was not tracked or triggered in time, 81% encounter discrepancies between invoiced and contracted prices at least occasionally, and 28% see them frequently or always. 94% run some automated invoice validation against contract terms; 46% have fully automated it.
Why it matters: negotiated savings that are never triggered, or are invoiced at the wrong price, do not reach the P&L, and the team reporting them is confident they did. Checking an invoice against contract price, dates and triggers is bounded, rule-based work with a definite answer, the kind of task an agent equipped with deterministic checks could plausibly take on.
What to do this week: pick your ten largest supplier contracts and list, for each, the rebates, price steps, remedies and renewal dates that have a trigger. Write down who tracks each trigger today and how. That list is the specification for an agent, built or bought.
Source: Agiloft, "Hidden Procurement Value Gap" research, Sept. 15, 2026 (2,005 respondents, fielded Aug. 5–20, 2026)
Story 4: The reason behind rising cash balances may be operational as opposed to caution
AFP's 2026 Liquidity Survey, underwritten by Invesco, was released June 16, 2026 (309 U.S. treasury professionals, fielded March 2026), a mid-year benchmark, not a this-week finding; AFP revisited it in an article on September 21. 46% of organizations increased U.S. cash balances over the prior 12 months, up from 38% the year before; 14% reduced them. AFP names improved operating cash flow as a major contributor, alongside regulatory, political and geopolitical risk, financing activity and tariffs. Bank deposits fell to 42% of short-term investment balances, the lowest since 2011. 75% of organizations maintain a written short-term investment policy. 17% do not review it at all, and 19% review it only every two to four years.
Why it matters: if the cash came from operations rather than caution, it will keep coming, and a policy nobody has reviewed since 2022 is deciding where it sits.
What to do this week: find out when your short-term investment policy was last reviewed. If the answer is more than two years ago, that is the action item, whatever the balance.
Source: AFP, 2026 Liquidity Survey (underwritten by Invesco), released June 16, 2026 (older benchmark, not a this-week finding); policy-review figures from AFP article, Sept. 21, 2026
Story 5: Finance talent shortage appears to be a long term trend
A May 2026 data point, not a this-week finding: Personiv's Hybrid Finance Workforce report, a survey of 203 finance and accounting leaders across 19-plus industries, found the average number of open finance and accounting roles per company had climbed to 17, up from five in 2025 and two in 2024. 63% of leaders now use AI and automation specifically to reduce the pressure to fill open roles, up from 23% a year earlier, and 94% use outsourced talent.
Why it matters: two, five, seventeen over three years is not a hiring cycle. The AI and outsourcing numbers say leaders have stopped waiting for it to reverse.
What to do this week: count your open finance and accounting requisitions today against a year ago. If the count is climbing, plan capacity (automation, outsourcing, role redesign), not only recruiting spend.
Source: Personiv, Hybrid Finance Workforce Report, published May 21, 2026 (203 respondents); figures confirmed via Accounting Today, June 3, 2026; older benchmark, not a this-week finding
The CFO Operating Brief is published by PCS Insight, LLC. Content is general operational guidance for finance leaders and does not constitute investment, legal, tax, or accounting advice.