USPTO's AI Patent Reset Under Squires Shows Promise Despite Examiner Resistance

Director John A. Squires has spent his first eight months at the USPTO making clear: artificial intelligence deserves patent protection, and the agency's previous hostility toward AI applications was counterproductive.

The evidence is mounting. Squires overturned a Patent Trial and Appeal Board decision that rejected AI model improvements as ineligible subject matter. Deputy Commissioner Charles Kim issued guidance in August 2025 reminding examiners to apply existing frameworks properly, not create new hurdles for software and machine learning claims. Most significantly, the November 2025 inventorship guidance rescinded the Biden administration's approach that scrutinised AI-assisted inventions under heightened standards.

The Pannu Pivot

The November guidance represents a complete reversal. Previously, patent examiners applied the Pannu v. Iolab Corp. factors to determine whether AI systems could qualify as joint inventors, a legally dubious exercise that inevitably failed. The new framework treats AI systems as tools, nothing more.

"No separate eligibility standard applies when examiners consider applications for AI-assisted inventions," the guidance states. Translation: stop making AI patents jump through extra hoops.

This matters because the old approach produced absurd results. Patent prosecutors were forced to craft inventorship declarations that downplayed AI contributions to satisfy examiners who misapplied joint inventorship doctrine to non-human entities.

But policy shifts at headquarters don't instantly change examination culture. IPWatchdog community feedback reveals examiner scepticism about the USPTO's own AI tools. One attorney reported that seven of ten AI-generated search results "missed the mark by miles." Another noted that "USPTO management has been repeatedly told by examiners that their AI search is ineffective and they don't seem to care."

This disconnect explains why only 76 of 3,200 available slots in the Automated Search Reference Notification programme had been filled by April. The USPTO waived the $450 petition fee and extended deadlines to boost participation, but practitioners remain unconvinced.

Litigation Pressure Mounts

While the USPTO recalibrates internally, courts are sharpening their scepticism of AI patent claims. The Federal Circuit's recent precedent requires concrete implementation details, not high-level algorithmic descriptions. Generic computing functions wrapped in machine learning terminology won't survive 35 U.S.C. § 101 challenges.

This creates a curious dynamic. The USPTO is becoming more receptive to AI patents precisely when district courts are becoming more aggressive about early dismissals on subject matter eligibility grounds. Patent prosecutors face pressure to draft claims that satisfy both USPTO examiners seeking technical specificity and federal judges suspicious of software patents disguised as AI innovations.

The numbers tell the story. More than a thousand AI patent lawsuits have been filed worldwide, with litigation "almost certain to continue to ramp up in 2026," according to recent analysis from major IP practices. Companies are conducting freedom-to-operate searches and building defensive patent portfolios because the alternative is expensive litigation.

Trade Secret Complications

Not every AI innovation belongs in the patent system. The Federal Circuit's ruling in Applied Predictive Technologies v. MarketDial reminded companies that trade secret protection requires actual definition and adequate security measures. AI companies training models on proprietary datasets face a choice: seek patent protection for specific technical improvements or rely on trade secret protection for broader methodologies.

The decision often depends on disclosure tolerance. Patents require full technical disclosure in exchange for twenty-year exclusivity. Trade secrets offer potentially unlimited protection but demand absolute secrecy and provide no defence against independent discovery or reverse engineering.

What Comes Next

Squires's policy reset creates opportunity, but it doesn't solve fundamental problems. The Supreme Court's Alice/Mayo framework remains hostile to abstract software concepts, regardless of AI branding. Patent prosecutors must demonstrate technological improvements that go beyond automation of existing processes.

The real test will come when these newly-approved AI patents reach federal court. Early motion practice on subject matter eligibility has become standard in software cases. District judges won't defer to USPTO examination if claims read as functional abstraction rather than technical invention.

Copyright litigation over AI training data continues separately, with publishers and content creators challenging fair use defences. These parallel proceedings will determine whether AI companies face liability on both ends: patent infringement for their technologies and copyright infringement for their training methods.

The USPTO's more accommodating stance creates breathing room for AI patent prosecution. Whether that translates to stronger patents that survive litigation remains to be seen.