USPTO cuts its patent backlog—but GAO says speed may be trading away examiners and quality

The story

The U.S. Patent and Trademark Office has cut its queue of unexamined patent applications by nearly 10% from its January 2025 peak, but a new Government Accountability Office audit says the faster pace may not be sustainable. The backlog fell from about 838,000 applications to 756,000 by August 2026. At the same time, patent-examiner attrition remained at 7.6%, applicant waits for a final patentability decision increased 11% from January 2025, and some quality controls were reduced or paused.

The November 5 report, GAO-27-108491, is not an argument for slowing every examination. It is a warning that the office’s strategy depends on a specialized workforce that takes years to build. GAO modeled utility, plant and reissue applications using fiscal 2020–2025 data and conducted 36 interviews with examiners, former USPTO officials, patent attorneys, companies, trade associations and researchers. Its central finding is that higher output cannot return the backlog below 599,000 before 2031 unless the agency also lowers attrition.

The workforce trend is stark. Overall examiner attrition rose from 2.9% in fiscal 2020 to 7.6% in fiscal 2025 and was still 7.6% on a trailing 12-month basis in August 2026. Training a junior examiner generally takes two to four years, yet about 40% of new examiners hired from fiscal 2021 through 2023 left during their first year. Hiring can add headcount, but rapid early departures consume training resources before those recruits develop the authority and experience of primary examiners.

The queue has nevertheless moved in the right direction. The USPTO said in April that first office actions during the fiscal year had begun exceeding new filings for the first time in nearly a decade. It reported meeting its statutory quality-compliance targets and highlighted progress on applications older than 36 months. GAO confirms the decline in unexamined inventory, but adds a second dashboard: average first-action pendency was 20.6 months in August, total pendency was 29 months, and most applications waited roughly 18 months just to be assigned to an examiner.

That distinction matters to inventors and investors. The backlog counts applications awaiting an initial decision, while total pendency runs until a patent is granted or the application is abandoned. A smaller intake queue does not automatically mean a faster final answer. GAO notes that long waits can impede fundraising and leave competitors uncertain about what protection may eventually issue. Delays can also trigger patent-term adjustments, shifting part of the administrative cost into longer exclusivity after grant.

GAO links the retention risk to specific management choices. The agency has required examiners to review more applications, shortened initial training and eliminated time assigned for some subject-matter training. Interviewees said higher production and timeliness requirements were encouraging departures and discouraging promotions into grades with still greater output expectations. The audit does not prove that every departure was caused by workload, but its forecast shows why attrition is not a separate human-resources issue: it directly changes examination capacity.

Artificial intelligence has not yet closed the gap. The USPTO has expanded AI-assisted patent searching and described the tools as increasingly robust. Stakeholders interviewed by GAO said the systems did not reliably return useful prior art, sometimes producing irrelevant results that examiners then had to verify. That is evidence about current operational experience, not a finding that AI search cannot improve. A useful tool must reduce search time while preserving recall across technical language, older patents and non-patent literature—not merely generate a longer candidate list.

Quality measurement is the other unresolved side of the equation. GAO says the office paused an external survey covering responsiveness and examination quality, stopped publishing some internal and external survey results, and reduced the use of quality reviews for examiner training. A 2025 GAO report had already found that examiner performance reviews could overstate compliance and that the office lacked a single goal for simultaneous compliance with all major patentability requirements. Removing feedback while raising throughput makes it harder to detect whether faster decisions are also weaker decisions.

The audit carries an unusual institutional caveat. GAO says the USPTO declined to provide requested information during the review because the agency disputed GAO’s authority to conduct work requested by congressional ranking members. GAO proceeded under the Comptroller General’s audit authority, shared a statement of facts in August and incorporated agency comments where appropriate. The USPTO did not comment on the final recommendation: that its director assess and address risks to examiner retention and examination quality. GAO lists that recommendation as open.

INNOVOX analysis: a patent office is not a document-processing factory. Each application asks an examiner to understand a technical claim, find relevant prior art and apply novelty, nonobviousness and eligibility rules. Production counts matter, but errors can grant rights that block legitimate competition or reject inventions that deserve protection. The durable solution is therefore a three-part system: retain trained examiners, measure quality independently and deploy search technology only when evidence shows it improves both speed and judgment.

What to watch next is whether the USPTO can keep the backlog falling while attrition drops below its recent level. Public data should show cohort retention, first-action and total pendency, reopened examinations, appeal outcomes and AI-search performance by technology area. Restoring training time and outside quality feedback may reduce short-term output, but could preserve capacity over several years. The new audit does not erase the office’s progress; it defines the conditions required for that progress to last.

INNOVOX analysis

The patent office has improved its visible queue while weakening some of the human capacity and quality signals needed to sustain that improvement. Faster throughput is valuable to inventors, but the system cannot treat each office action as interchangeable output: a weak search or rushed decision can shift much larger costs into appeals, litigation and uncertain investment.

What to watch

Watch the 12-month examiner attrition rate, first-action and total pendency, the retention of newly hired examiners, restoration of subject-matter training and quality surveys, and measured precision of AI prior-art tools. The decisive policy signal will be how the USPTO responds to GAO’s open recommendation without reversing the recent backlog decline.