Greenberg Traurig | Recalibrating AI Patent Strategy: What the USPTO Shift Means – and What it Doesn’t

Under new leadership, the United States Patent and Trademark Office (USPTO) has recently adjusted its approach to patentability of artificial intelligence (AI) technologies in multiple meaningful respects. Many companies have not yet fully absorbed the practical implications of those shifts. If your understanding of what is or is not patentable in the AI space was formed under USPTO practices from even a year ago, that understanding is worth revisiting. Companies developing and deploying AI technologies should be assessing whether those shifts create protection opportunities that did not previously exist, or whether prior decisions about what to file or how to claim it deserve a second look.

That assessment, though, is not as simple as it might appear. The USPTO operates within a legal framework established by Congress and interpreted by the courts, and the courts have not changed course. Making sound decisions about how to leverage the new USPTO options, while ensuring consistency with the law that courts will apply, requires care and experienced (intellectual property) IP counsel to navigate both systems and pursue protection that is not just granted by the USPTO, but is valid and enforceable.

Recalibrating what you thought you knew

The adjustments at the USPTO have arrived through a series of internal guidance documents and administrative decisions issued under new Director John Squires, who was nominated in early 2025 and formally took office in September 2025. A memo to the patent examining corps from August 2025 emphasized limitations on the “mental process” grouping, a frequent basis for rejecting AI-related claims on the theory that the method could be performed in a human mind. Machine learning algorithms that process large volumes of data and perform calculations well beyond human cognitive capacity, the memo reasoned, should not be categorically swept into that exclusion. Soon after, an Appeal Review Panel decision vacated a Board rejection of machine learning claims and accentuated patentability of improvements in operations of machine learning models. Most recently, the USPTO released “Subject Matter Eligibility Declaration” memos in December 2025 that introduced a procedural mechanism allowing applicants to submit objective evidence and expert testimony in support of patent-eligibility. Each of these developments reflects an intent to prevent patent laws from functioning as an automatic disqualifier for AI-related innovations.

For companies and counsel whose AI patent strategy was shaped by the USPTO’s prior disposition, which often necessitated extensive prosecution to achieve broad claims, these adjustments are worth factoring in. Technology that may have been deemed too abstract to pursue in prior years could now present a more viable path to protection. Claims that previously required significant workarounds to satisfy examiner concerns may be better positioned under new guidance.

This is particularly relevant for companies whose counsel previously advised against filing certain categories of AI applications due to eligibility concerns, or who pursued narrower claims than the underlying technology might have supported. Against this backdrop, it is reasonable to revisit prior filing decisions and assess whether the updated examiner guidance around the improvement-to-technology framework creates space for claims that were not practicable before. The emphasis in recent USPTO materials on specifications that expressly describe how a claimed invention improves the machine learning model itself, or a related technical field, can inform how new applications are structured going forward.

But consider the courts

While the USPTO presents new options, the courts present something of a counterpoint. District courts, and the Court of Appeals for the Federal Circuit that hears patent appeals, continue to apply the same Alice/Mayo Supreme Court two-step analysis with the same basic architecture in use for years. Companies that read the USPTO’s adjusted posture as a signal that AI patents are now broadly secure should take care to temper their enthusiasm.

An April 2025 Federal Circuit decision on AI technologies illustrates where the legal floor continues to sit. The court affirmed the invalidity of four patents directed to the use of machine learning for generating broadcast network maps and event schedules. The court found that the patents merely claimed a generic application of machine learning techniques in the entertainment scheduling context. Underlying the court’s decision were its findings that the claims at issue did not describe how the machine learning model itself was improved, what specific implementation distinguished the approach from conventional techniques, or how the technology advanced. Merely applying established machine learning methods to a different field, the court held, does not create a patent-eligible invention.

The case serves as a reminder that the Federal Circuit’s doctrinal framework has not moved in parallel with the agency’s administrative stance. A patent obtained under the more receptive examination environment that currently exists may still face invalidity challenges in litigation governed by precedent that has not softened. The gap between what the USPTO will allow and what the courts will sustain is a real risk, and one that patent strategy must account for.

A note on the UK

The United States is not the only jurisdiction where the treatment of AI inventions has recently shifted. In February 2026, the UK Supreme Court issued a significant ruling holding that an artificial neural network-based recommendation system was not excluded from patentability as a program for a computer as such. In doing so, the court abandoned the longstanding framework that had governed computer-implemented inventions in the UK for nearly two decades and aligned UK law with the European Patent Office’s approach. The practical effect is a lower eligibility threshold in the UK, with assessment moving more squarely to novelty and inventive step. Companies managing international IP portfolios should consider the new filing opportunities that did not exist under prior UK practice.

Where to go from here

For many companies pursuing IP to support their businesses, their objective in this new landscape will not simply be to take advantage of a more favorable examination environment. It is to do so in a way that produces patents capable of withstanding the scrutiny that courts continue to apply. Understanding the gap between what the patent offices will allow and what the courts will sustain, and structuring prosecution strategy to account for it, is where experienced IP counsel adds meaningful value.

For companies developing AI technologies, that means claim drafting and specification development require more precision, not less. A patent application that may now progress more smoothly through examination, such as one describing an AI system that improves a technical process, must still be grounded in specificity sufficient to survive a Section 101 challenge. The Federal Circuit’s concerns in the case noted above were that the claims were functional and broad, describing outcomes rather than mechanisms. Specifications should be drafted to articulate, in concrete technical terms, the specific improvements to a model’s architecture, training approach, or output mechanism that distinguish the invention from simply applying existing machine learning techniques in a new setting. The Subject Matter Eligibility Declaration procedure, while a useful addition to the practitioner’s toolkit, reinforces the same point. These declarations may strengthen prosecution and support issued patents against future challenges, but they are not a substitute for applications built on technical specificity from the outset.

Developing an effective strategy requires detailed discussions with IP counsel experienced in both the technical substance of AI innovations and the legal framework governing their protection. The current USPTO environment presents real opportunities for companies that approach it deliberately, with counsel who can identify which technologies and claim structures are well-suited to the new examination landscape and which remain exposed under prevailing court doctrine. Companies that treat these changes as a general invitation to file broadly, without that precision, may find themselves with a portfolio that looks strong until it is tested.

Compliments of Greenberg Traurig – a Premium Member of the EACCNY