The September 2026 Employment Situation, released by the US Bureau of Labor Statistics on 2 October, showed payroll employment up 29,000 and the unemployment rate at 4.2 percent. The bureau said both changed little, and that employment in every major industry changed little over the month. Private forecasts had looked for something nearer 84,000 to 90,000. That gap is a miss against the surveys. It is not, on the bureau’s own wording, evidence that artificial intelligence removed the jobs.
Unemployed people numbered 7.1 million, also little changed. The rate has stayed between 4.1 and 4.3 percent since March. Table A shows 4.1 percent in August and 4.2 percent in September. That tenth of a point is not a jump, and it is not a count of roles automated away.
What the bureau reported
Summary Table B prints September at +29,000, preliminary. Private employers added 46,000. Government fell by 17,000. Average hourly earnings were $37.81, up 5 cents on the month and 3.0 percent over the year. Heather Long of Navy Federal Credit Union, via CBS, called that the lowest annual gain since May 2021. The 3.1 percent figure is August’s earlier reading, not this print.
Long also said, via CBS: “Only healthcare and construction were hiring, and it was weak.” That “only” is hers. The release says health care added 17,000, about half a prior-12-month average of 33,000, and manufacturing added 9,000 and is up 72,000 since a low in December 2025. Table B’s health-care-and-social-assistance line is +23,000. Construction added 11,000. Information fell 10,000 and professional and business services fell 9,000, which the narrative still calls little changed. Financial activities fell 7,000.
The release marks the prior two months down by 60,000. July went from +21,000 to −10,000. August went from +162,000 to +133,000.
On 16 September the Federal Open Market Committee voted 12–0 to raise the funds-rate target by a quarter point, to 3.75–4.00 percent. ABC and CBS called it the first increase in three years. The print does not set October.
What forecasters had in hand
These were separate surveys. They are not a Bureau of Labor Statistics forecast, and they are not one consensus.
September 2026, forecast against the print
Data figure
Forecasts, then the print
Forecasts published before the print, and the Bureau of Labor Statistics figure. Not one poll.
Source: FactSet, 1 October 2026; ABC News, 2 October 2026; MUFG, 1 October 2026; US Bureau of Labor Statistics, 2 October 2026.
| Item | What was forecast | September print, 2 October |
|---|---|---|
| Nonfarm payrolls, FactSet | Median +90,000. John Butters, 1 October. 27 estimates, from 60,000 to 130,000. | +29,000. About 61,000 below that median. The bureau said the change was little. |
| Nonfarm payrolls, ABC News | Economists expected +84,000. | +29,000. About 55,000 below that expectation. |
| Nonfarm payrolls, MUFG | Own forecast +90,000, 1 October. It also cited a Bloomberg contributor median of +85,000, and a range of about 70,000 to 110,000. | +29,000, below the bottom of that range. |
| Unemployment rate | MUFG expected 4.1 percent to hold, with a risk of 4.2 percent. | 4.2 percent. Table A shows 4.1 percent in August. The bureau said 4.2 percent changed little. |
| Private payrolls | ADP, a different survey, +90,000. CNBC, 30 September. | Bureau private payrolls +46,000. The two surveys are not a score against each other. |
| Average hourly earnings | Not a FactSet figure in the notes used here. | $37.81, up 5 cents on the month and 3.0 percent over the year. |
| Financial activities | No forecast in the notes used here. | −7,000. |
The release says September followed an average monthly gain of 45,000 over the prior 12 months. FactSet, on 1 October, had used a trailing 12-month average of 50,300, before this report and these revisions. Keep 50,300 as FactSet’s pre-release figure.
August 2026, forecast against the first print and the revision
August was not a miss. The first print beat the FactSet median. The 2 October revision lowered it and it was still above that median.
| Item | Forecast before the August release | First published | After 2 October |
|---|---|---|---|
| Nonfarm payrolls | FactSet median +65,000. John Butters, 3 September. 6 estimates, from 50,000 to 70,000. | +162,000. Bureau of Labor Statistics Economics Daily, 10 September. About 97,000 above that median. | +133,000. Summary Table B, preliminary. About 68,000 above the 3 September median. |
| FactSet’s later comparison | On 1 October FactSet set the unrevised 162,000 against a median of 60,000. That is not the 65,000 median from 3 September. | +162,000 versus 60,000. | Leave 133,000 off this row. The 1 October note predates the revision. |
| July payrolls | — | +21,000, the earlier figure named in the September release. | −10,000. Revised down by 31,000. With August’s 29,000 markdown, the two months are 60,000 lower than previously reported. |
| Unemployment rate | Not in the FactSet payroll note used here. | — | 4.1 percent in August. Table A. |
Where the monthly changes sit
Data figure
September 2026, over the month
The bureau said employment in every major industry changed little.
Diverging monthly change by industry, in thousands. The chart is composition, not a league table of industries “exposed to AI”.
Source: US Bureau of Labor Statistics, Summary Table B, September 2026, preliminary.
The chart is composition, not a league table of industries “exposed to AI”. Gains in health care and social assistance, construction, and manufacturing sit next to losses in information, professional and business services, and financial activities, inside a month the bureau called little changed. A weak print is not replacement. The longer argument is on what a layoff count does not prove about AI.
Finance and insurance, without a cause tag
Financial activities fell by 7,000 in September and are down 129,000 since a peak in May 2025. The release says most of that loss is insurance carriers and related activities, −90,000.
Through August, not this Friday, Insurance Business on 9 September reported finance-and-insurance employment of 6,651,100, down from 6,733,400 a year earlier, about 82,000, with carriers down 6,300 in August alone. It called the cause murkier than a model purchase. That piece predates the 16 September rate rise, so it is not a reading of Friday. A sector total is not a count of marketing roles, and the bureau does not tag a separation as caused by a tool.
Claims work, where the occupation data is sharper
The sharper occupation evidence is claims adjustment, and it is older than Friday. From June 2025 to May 2026, Glassdoor found 98 percent of claims-adjuster comments that mentioned AI were critical, against 53 percent across all jobs, and 81 percent of insurance comments that mentioned AI were negative. Claims Journal carried that on 27 August. One review said people spent 15 to 20 minutes correcting an AI draft a supervisor was not checking. That is a complaint about an unsigned draft, not a job count.
Entry-level adjuster postings were down close to 50 percent since early 2024, against about 15 percent for entry-level jobs overall. Senior postings were about 80 percent above 2017. All adjuster postings were about 55 percent below the post-pandemic peak, against about 36 percent for the wider market. Claims Journal said the occupation lost 13,100 jobs in the past year, a large portion of 74,800 in the broader sector. A separate comparison, in Insurance Business on 28 August: employment in the claims-adjusting industry was down about 21 percent in the year through May 2026, against 2.5 percent for insurance carriers and related activities.
Data figure
Decade outlook, not this year
Projected employment change, 2025–35. This is not the past year’s change in claims jobs, and it is not a forecast from Integrated.Social.
Source: BLS Occupational Outlook Handbook: claims occupations; insurance underwriters; bookkeeping clerks; accountants and auditors.
From 2025 to 2035 the handbook projects claims adjusters, examiners, and investigators down 5 percent. The broader group, including auto-damage appraisers, is down 6 percent, about 21,800 jobs, from 389,700. Insurance underwriters are down 4 percent. Bookkeeping, accounting, and auditing clerks are down 6 percent. Accountants and auditors are up 5 percent. Insurance Business noted about 21,600 openings a year even on that decline, because people leave. A decade projection is not September’s print.
Acrisure did name the technology. Insurance Journal, on 22 May 2026, reported that Greg Williams told staff about 2,250 roles, about 11 percent of roughly 19,000, mostly in the United States, would go in phases from 21 May into 2027. He tied it to technology, AI, and digital platforms. A consultant on the same page cited slow organic growth and debt. It is an announced plan, not a finished cut, which is the point of AI-washing and human expertise.
Investment banking: a column, not a headcount table
This is a change in the work, not a closed analyst programme. Reuters Breakingviews, on 29 June 2026, said big banks and boutiques are generally not planning to shrink analyst recruiting in a material way, from conversations with managing directors and others familiar with plans at eight banks. A handful were hiring more. Models and slides were not reliable enough for a managing director to drop the junior check. Rainmakers were already using the tools for meeting prep and earnings summaries. It is a column, not a headcount table.
On Lazard’s 23 July call, the smaller-team line is chief financial officer Tracy Farr’s: smaller deal teams and a lower associate-equivalent per managing director. Peter Orszag said the firm had turned over 40 percent of advisory managing directors, is still adding a net 10 to 15 a year, and is hiring in healthcare, industrials, and technology, plus one defense-tech hire, disproportionately in the United States. Advisory managing directors were 238 at the end of the first quarter, with a target of at least 248 a year later. AI spend was increasing and not yet material. A wrong number on a live deal is still a person’s problem. Judgement over volume is the argument on what to hire for when volume is cheap.
Two Goldman tools, not one rollout
Comparison
Two Goldman tools, not one rollout
| GS AI Assistant | Banker Copilot | |
|---|---|---|
| What it is | Firmwide assistant | Specialist tool for bankers, still being refined for broader use |
| Scale you can cite | Weekly use by about 40,000 of 46,200 employees, millions of questions a month (TIME, 2026). Earlier: over one million prompts a month (Observer, 18 September 2025). | No public user count and no public productivity figure. |
| What not to say | That it runs without a person. | That it is the same product, that it has left a pilot, or that a marketing team can copy it as “the bank already did this”. |
Sources: TIME, 2026; Observer, 18 September 2025.
Banker Copilot and the GS AI Assistant are different products. Argenti told the Observer on 18 September 2025 that Banker Copilot “is currently not scaled, but is being leveraged by a select group who are optimizing it before it can be deployed more broadly.” Business Insider, on 13 May 2025, put that access in the dozens. No 2026 story checked for this piece says it has left the pilot, and there is no public productivity figure for it.
The assistant is the firmwide tool. That same 2025 interview already had it above one million prompts a month. TIME’s 2026 profile is the later scale: weekly use by about 40,000 of 46,200 employees, millions of questions a month, and more than 100 specialised versions. More than 12,000 engineers had a coding copilot. Argenti put those gains at at least 20 percent. Business Insider’s “up to 20 percent,” in May 2025, was Melissa Goldman, on the developer tool.
A third effort, in February 2026, was Anthropic work on trade accounting and client onboarding. Argenti told CNBC it was early, that job losses there would be premature, and that pitchbooks were a possible next task, not a live tool. Axios, on 8 September 2026, reported that the systems do not get direct access to underlying data. That limit is the Authority Envelope. Argenti’s list in the Observer was describe, delegate, and supervise. That is the same point as the expert who still checks the draft.
What Friday does not show
CNBC, on the morning of 2 October, reported Draup figures that cut the other way. AI-related postings at JPMorgan Chase, Citigroup, and Capital One were up 49 percent versus 2025, to 139,819 listings. Agent-orchestration mentions were up 1,721 percent. Vijay Swaminathan called it “arguably the hottest skill on Wall Street.” Those are listings, not headcount, and they do not offset the decline in financial activities. The bureau still said the headline and every major industry changed little.
What this print is, and what it is not
| What the print supports | What it does not support |
|---|---|
| Payrolls +29,000 against a FactSet median of 90,000 and an ABC expectation of 84,000. The bureau said the change was little. | A collapse in every sector. |
| Unemployment 4.2 percent, little changed, inside 4.1 to 4.3 percent since March. | Evidence that AI removed the jobs. |
| Financial activities −7,000 in the month, and −129,000 since May 2025, most of it insurance carriers (−90,000). | A count of marketing roles, or proof a tool caused the loss. |
| Eight banks generally not planning a material cut in analyst recruiting (Breakingviews, 29 June). Lazard’s smaller-team line is Tracy Farr’s (23 July). | A closed junior bench, or that sentence in Peter Orszag’s mouth. |
| Banker Copilot is still a select-group pilot. The GS AI Assistant is the firmwide product. | One rollout a marketing team can copy. |
FAQ
Did AI explain the September 2026 US jobs report?
No. Payrolls rose 29,000 and unemployment was 4.2 percent, and the bureau said both changed little. Employment in every major industry changed little. A miss against a survey is not a count of jobs a model removed.
Did AI cause the losses in finance and insurance?
Not on this release. Financial activities fell by 7,000 in September and are down 129,000 since May 2025, most of that in insurance carriers and related activities (−90,000). The bureau still said every major industry changed little. The bureau does not tag a separation as caused by a tool. Through August, before this print, Insurance Business reported finance-and-insurance employment down about 82,000 on the year, to 6,651,100, and said the cause was murkier than a carrier buying a model.
Where in finance is the AI evidence actually sharp?
Claims adjustment, and the research is from August, not from this jobs report. Glassdoor found 98 percent of adjuster comments mentioning AI were critical, against 53 percent across all jobs. Employment in the claims-adjusting industry was down about 21 percent in the year through May 2026, against 2.5 percent for carriers and related activities. Entry-level adjuster postings were down close to 50 percent since early 2024. That does not prove the −7,000 in financial activities, and it does not prove a trading-floor cut.
Are banks cutting jobs because of AI?
Some say so. Acrisure’s May 2026 memo tied about 2,250 roles, about 11 percent by the end of 2027, to technology, AI, and digital platforms. The same coverage also named slow growth and debt. At JPMorgan Chase, Citigroup, and Capital One, Draup figures shared with CNBC on 2 October showed AI-related postings up 49 percent this year, to 139,819, and agent orchestration mentions up 1,721 percent. That is hiring into control of the tools, not a layoff table, and it does not offset the sector.
Are investment-banking analyst programmes being emptied?
No. Reuters Breakingviews, on 29 June 2026, said big banks and boutiques are generally not planning a material cut in analyst recruiting, from conversations about eight banks, and that a handful were hiring more. Lazard’s thinner-team line on the 23 July call is Tracy Farr’s, not Peter Orszag’s. Fewer juniors is not the same as no signer.
Is Goldman’s Banker Copilot replacing analysts?
No. On 18 September 2025 Argenti told the Observer it was not scaled and was with a select group. Business Insider in May 2025 put access in the dozens, and no later story checked here says that has changed. It is not the GS AI Assistant, which TIME in 2026 described as in weekly use by about 40,000 of 46,200 employees. There is no public productivity figure for Banker Copilot. Argenti told CNBC in February 2026 that it was premature to expect job losses from the separate Anthropic work on accounting and onboarding. A person still checks the output.
Was September a large miss, and was August?
September was, against the surveys. FactSet’s median was +90,000 and the print was +29,000. ABC said economists expected +84,000. August was the opposite on the first print: FactSet’s 3 September median was +65,000 and the bureau first published +162,000. The revision to +133,000 is still above that 65,000 median. August was not a miss.
Did unemployment jump from 4.1 percent to 4.2 percent?
Do not call it a jump. Table A shows 4.1 percent in August and 4.2 percent in September. The release said 4.2 percent changed little, and that the rate has stayed between 4.1 and 4.3 percent since March. About 7.1 million people were unemployed.
Should a marketing team change paid spend because of this print?
No. The claim still has to work in the ad and on the page. Tools may draft the variants. A named person still signs. Do not tie a budget change to this release, and do not add another login because a bank has a pilot.
Bottom line
September’s 29,000 was a large miss against FactSet’s 90,000 and ABC’s 84,000. August’s 162,000, later 133,000, was still a beat of FactSet’s 65,000. The bureau said September changed little. The evidence that names AI is narrower and earlier: claims work, and one broker’s announced plan. Bank listings for AI skills do not offset the finance decline, and none of it replaces the person who signs.
About the author
Modi Elnadi is Head of AI Marketing Growth and SEO/AEO at Integrated.Social. Agents draft. A person signs. This note uses the bureau’s tables and the named articles above. It is not a forecast of the next payroll print, and it is not a claim that a model can publish the reading.
