The Probationary Federal Workforce Fell 30%. Most Separations Were Classified as Voluntary.

The federal government's conditional workforce became much smaller during 2025. Across 11 large agencies selected by the Government Accountability Office for their health and safety missions, the number of employees in OPM tenure group 2 fell from 332,928 at the end of 2024 to 231,614 one year later.

That was a decline of 101,314 employees, or 30.4%. Total employment at the same agencies fell 8.7%.

The difference is evidence that the reduction reached the government's entry-and-development pipeline disproportionately. It is not evidence that every person in the group was a recent hire, that every departure was a firing, or that every separation created a lasting vacancy.

Those distinctions matter. The cohort includes career-conditional employees and people serving appointments that require probationary or trial periods. GAO cautioned that the public data cannot separate employees who were newly hired from those in a probationary period after appointment as supervisors or managers. It also found that 78.6% of the cohort's 2025 separations were classified as voluntary, including retirements and the deferred resignation program. Involuntary actions such as terminations and reductions in force accounted for 17.8%; transfers made up the remainder.

The more defensible conclusion is narrower and more consequential: the government ended 2025 with far fewer employees in the status from which much of its future career workforce develops. The effect was highly uneven. Transportation Security Administration probationary and trial employees separated at more than four times the rate of TSA's workforce overall. At the National Park Service, the cohort's separation rate was far lower than the agencywide rate.

A probationary employee is not necessarily a first-year employee

The terminology is easy to flatten. In the competitive service, an initial probationary period generally follows a career or career-conditional appointment. In the excepted service, a similar stage is called a trial period. Federal rules also require probationary periods for some first appointments to supervisory or managerial positions.

For its June 2026 review, GAO used OPM tenure group 2 to identify the population. OPM defines that group to include career-conditional employees, career employees serving initial-appointment probation, employees serving trial periods, and equivalent conditional appointments in the excepted service. The group is a useful measure of employment that has not reached unrestricted career tenure. It is broader than "people hired during the past year."

GAO selected 11 parent agencies and 12 subcomponents whose missions include air travel, disaster response, food and public health, national and nuclear security, natural resources, transportation, and veterans' care and benefits. The parent and component rows overlap. The Forest Service is inside Agriculture, for example, and the Veterans Health Administration is inside Veterans Affairs. They can be compared, but they cannot all be added together.

Across the 11 non-overlapping parents, total staffing declined from 1,881,582 at the end of 2024 to 1,717,996 at the end of 2025. The probationary/trial cohort fell three and a half times as fast in percentage terms. Its share of selected-agency employment dropped from 17.7% to 13.5%.

GAO calculated separation rates by dividing 2025 separations by the average of beginning- and end-of-year employment. That method accounts for a population that changed substantially during the year. It found 52,734 separations from the parent-agency cohort, equal to 18.7% of its average probationary/trial workforce. The same calculation produces 14.9% for all employees.

The table resists a single narrative. The cohort separated faster than the overall workforce at 18 of the 23 parent and component rows, but the size of the gap ranged from almost nothing at Defense, FAA and EPA to 35.8 percentage points at TSA. Four rows moved in the opposite direction: Interior, the Bureau of Land Management, the National Park Service and the Nuclear Regulatory Commission.

The workforce decline was broad, but it was not uniform

The headcount change and the separation rate answer different questions. A separation rate measures the flow of exits relative to the average population. The year-end change also reflects accessions, transfers into the group, changes in tenure and any rehiring. GAO noted that some employees may have been rehired, but the data do not say how many.

Among the 11 non-overlapping parents, every probationary/trial population except Homeland Security's was smaller at the end of 2025. Energy's cohort fell 52.1%, HHS's 43.3%, Agriculture's 45.9%, Commerce's 41.4% and Veterans Affairs' 41.5%. Defense lost the largest number, 44,703, but its much larger base made the decline 27.4%.

Homeland Security was the major exception. Its group ended the year essentially unchanged, rising by 81 employees, even though it recorded 6,039 separations. That combination is possible only because employees were also entering or moving into the group. Public aggregate data do not establish which hiring programs or components accounted for the replacement.

The National Park Service provides another warning against treating a parent department as a proxy for every component. Interior's overall separation rate was 30.9%, and NPS's was 45.9%, reflecting the park system's large seasonal workforce. The NPS tenure-group-2 rate was 15.2%. Seasonal employees can be in other appointment and tenure categories, so the agencywide rate and the conditional-career rate describe different parts of the workforce.

TSA shows the opposite pattern. Its overall workforce changed from 65,805 to 63,006, a decline of 4.3%, while its probationary/trial population fell from 7,758 to 4,533, or 41.6%. GAO recorded 2,890 cohort separations, yielding the study's highest rate at 47.0%.

These contrasts are not proof that one agency protected its talent pipeline while another deliberately dismantled it. They show that departmentwide totals conceal appointment structures, seasonal work, continuing hiring and component-specific actions.

Most separations were classified as voluntary

The words used for 2025 departures have often been imprecise. GAO grouped resignation, retirement and deferred resignation as voluntary. It grouped terminations and reductions in force as involuntary. Transfers were separate.

Using the public extract available for this analysis, FederalHiringData counted 55,507 tenure-group-2 separation actions across the same parent agencies. The total is slightly higher than the 52,734 in GAO's published table, consistent with a later public-data version; this analysis does not infer the cause of the revision. The current extract includes 42,399 quits, 690 voluntary retirements, 191 early-out retirements, 91 other retirements, 3,409 termination actions, 976 RIF actions, 5,625 other separation actions and 2,126 transfers.

When those codes are placed into GAO's broad groups, the current extract is closely aligned with the report: 78.1% voluntary, 18.0% involuntary and 3.8% transfers.

"Voluntary" is an administrative classification, not a finding about why an employee chose to leave. The deferred resignation program allowed participating employees to agree to resign or retire later while remaining on paid administrative leave. A coded quit does not reveal whether the person welcomed the choice, saw little opportunity to remain, had another job, or responded to organizational uncertainty.

The reverse caution also applies. An involuntary action is not necessarily a performance-based firing. OPM's category for termination includes expired appointments and other actions; GAO's discussion also includes RIF. The available aggregate data cannot attribute individual motives, performance, legal basis or later employment.

The policy environment nevertheless changed materially. Executive Order 14284 directed agencies to make an affirmative public-interest determination before finalizing covered appointments and allowed consideration of agency needs and organizational goals in addition to performance and conduct. OPM's June 2025 final rule implemented related changes. GAO documented the legal challenges and later appeal-rule proposal; this article does not resolve those legal questions.

Younger employees had higher rates, but age is not tenure

The group was not confined to young workers. At the end of 2024, 36% of the selected parent-agency cohort was age 40 or older. At the end of 2025, the share was 43%. That shift occurred because the younger counts fell faster, not because the older counts grew.

Separation rates followed a U-shaped pattern. The rate was 23.5% for ages 20 through 24 and 21.5% for ages 25 through 29. It declined to 16.2% for ages 50 through 54, then rose to 27.9% at age 65 or older, where retirement accounts for more departures.

The age pattern supports concern about an early-career pipeline, but it cannot quantify one by itself. A 30-year-old can be an experienced federal employee entering a new appointment. A 55-year-old can be a first-time federal hire. A newly appointed supervisor may have decades of government service. Age, tenure group and length of federal service are related but not interchangeable.

The action mix also weakens a simple youth-termination story. Among cohort members ages 20 through 29, the current OPM extract records 15,596 separations; 12,213, or 78.3%, were in the broad voluntary group. Involuntary actions accounted for 2,932, or 18.8%, and transfers for 451.

What the age data establish is exposure: the conditional workforce lost a large number of employees in the age bands from which agencies often build long careers. They do not establish whether those individuals would have stayed for decades, whether their skills were scarce, or whether their work was replaced.

Inspection, IT, contracting and natural-resources occupations were involved

The largest occupation-series count was compliance inspection and support, series 1802. Across the selected parent agencies, the tenure-group-2 cohort recorded 3,417 separations in that series and a 38.1% separation rate. Miscellaneous administration and program recorded 2,394, management and program analysis 2,195, and information technology management 2,115.

Clerical and assistant work, forestry technicians, contracting, business and industry, natural-resources management, medical support, human resources, engineering, police work, nursing and public-health occupations also appear among the largest counts.

An occupation series is not a job assignment. A series 2210 employee may work in cybersecurity, application development, infrastructure or program management. A series 1802 employee can support different inspection and enforcement missions. The counts do not show which skills left, whether contractor capacity replaced them or whether a mission metric changed.

They do show why succession planning cannot be reduced to an agencywide percentage. The affected population included the occupations that inspect compliance, procure goods and services, maintain information systems, manage natural resources and support medical operations. Rebuilding those capabilities can require security clearances, specialized credentials, institutional knowledge and supervised development after hiring.

The geographic footprint was incomplete and uneven

GAO also published state-level counts for selected agencies, but OPM redacted location data for Defense, Homeland Security and employees in sensitive or national-security occupations. The published geography excluded 191,643 employees for those reasons and another 1,427 records with no reported location.

Among the remaining records, Montana had 525 separations against 1,667 employees in January 2025, a ratio of 31.5%. Idaho was 31.1%, Alaska 29.0% and Utah 25.2%. Maryland had the largest reported separation count, 2,585, and a ratio of 23.9%. California recorded 1,924 separations, an 18.2% ratio; Texas recorded 1,223, or 15.8%.

These ratios are not the same measure as GAO's agency separation rates. The location table provides a January workforce count, not the average of the beginning and end of the year. The chart therefore divides separations by the January count and labels that denominator explicitly. It is useful for showing where reported departures were concentrated, not for ranking states on a fully comparable national rate.

Related research

Put this finding in context

All articles

Data investigations/August 24, 2026

IRS Had 126 AI Use Cases. Its Technical Workforce Fell 28%.

Most IRS AI use cases were still in development as a selected technical-workforce proxy fell by 2,459 employees from January 2025 through June 2026.

Data investigations/August 21, 2026

VHA's Federal Workforce Fell by 28,342. Its Care Forecast Kept Rising.

VA's health workforce contracted after late 2024 even as patient, visit and community-care demands remained large. The losses reached clinical and coordinating roles.

The omissions are large enough to rule out claims about the entire federal map. They also mean that the geography cannot describe TSA's 47% rate or Defense's 20,204 cohort separations. Those are major pieces of the national total.

Recruiting had built a large pipeline before the contraction

The FederalHiringData historical USAJOBS archive recorded 312,791 distinct announcements opened during 2024 across the selected departments and the Army, Navy and Air Force. The same bounded group recorded 289,593 in 2018, 299,590 in 2019, 276,012 in 2020, 317,886 in 2021, 377,893 in 2022 and 392,639 in 2023.

An announcement is not a hire or a vacancy. One control number can cover multiple locations, grades or positions, and an agency can cancel an announcement or hire fewer people than it anticipated. The archive does not produce an applicant-to-hire conversion rate.

The series instead establishes the scale of public recruiting that preceded the 2025 reduction. Agencies were not drawing their conditional workforce from a small or static posting system. They had spent years advertising hundreds of thousands of opportunities annually across the selected departments.

Longer OPM history provides another reference point. Government-wide tenure-group-2 headcount rose from 266,810 in December 2015 to 418,743 in December 2024. It fell to 278,010 in December 2025 and 240,539 in June 2026. The June figure is a partial-year snapshot, not a full-year comparison, and the public series available in the local warehouse does not support a comparable extension to 2000 for this specific tenure category.

That longer decline does not reveal how many separated employees were rehired, how many moved into tenure group 1, or how many new employees entered after the snapshot. It does show that the conditional population remained well below its pre-reduction peak into mid-2026.

The measurable risk is a thinner development pipeline

The evidence supports a workforce-pipeline concern without proving a service failure. Agencies need a continuous supply of people who can move from conditional appointments into experienced career roles. They also need supervisors and managers capable of developing those employees, transferring institutional knowledge and certifying readiness for continued service.

The 2025 change reduced that pool sharply. It reached IT, contracting, inspection, enforcement, health, engineering and land-management occupations. It was especially pronounced at some agencies and modest or reversed at others. Most departures were not recorded as involuntary, and the public files cannot tell whether a particular employee's skills disappeared from government or were later replaced.

The next useful measures are therefore concrete: cohort retention by occupation, conversion into unrestricted career tenure, time to replace separations, training completion, vacancy duration and mission-specific performance. Public aggregate data currently answer only parts of that chain.

Calling every separation a firing would overstate the evidence. Treating a 30% contraction in the conditional workforce as ordinary turnover would understate it. The federal talent pipeline became materially thinner in 2025. Whether agencies can rebuild it without losing critical capability is the question the next rounds of workforce and performance data must answer.

Methodology and limitations

FederalHiringData reproduced GAO-26-108557 table 1 for all 11 selected parent agencies and 12 selected subcomponents. Parent and component rows overlap and are never added together. Aggregate totals in this article use only the 11 parent rows. GAO calculated separation rates using the average of Dec. 31, 2024 and Dec. 31, 2025 workforce counts.

The article also analyzed OPM Federal Workforce Data and Enterprise Human Resources Integration public files in the FederalHiringData research warehouse. The cohort uses tenure code 2, matching GAO's stated method. OPM action categories were grouped to align with GAO's voluntary, involuntary and transfer presentation. The current public extract contains 55,507 selected-parent actions, slightly more than GAO's published 52,734; the article treats the official GAO table as authoritative for the report-period totals and labels current-extract calculations separately.

Age and occupation rates divide 2025 separations by the average of December 2024 and December 2025 headcount. Occupation analysis is limited to series with an average cohort of at least 500. Age is not length of service, occupation is not assignment, headcount is not full-time-equivalent labor, and separation does not identify a vacancy or mission effect.

The location analysis uses GAO table 2. Its denominator is January 2025 headcount, not an average workforce. DOD, DHS and other sensitive records are redacted, and some locations are missing. It should not be read as a complete national ranking.

The recruiting analysis counts distinct USAJOBS control numbers opened in full calendar years 2018 through 2024 and attributed to the selected departments or military departments. Historical archive coverage begins in March 2017, so 2017 is excluded. Announcements are not hires, vacancies or employees.

GAO said its data cannot distinguish employees newly hired into government from those serving a probationary period after appointment as supervisors or managers. The data also do not identify how many separated employees were rehired. This analysis does not infer motive, performance, causation or future mission outcomes. Sources and current policy pages were checked through Aug. 24, 2026. Readers can browse current federal openings, review federal hiring statistics, explore agency profiles, or read more FederalHiringData investigations.