AI patents have become easier to obtain over the past year. Where an issued AI patent can be tested for eligibility has not changed, and that constraint decides more cases than any other.

A negative subject matter eligibility decision ends more AI patent cases than any other single ground of invalidity. It is also the ground that most administrative tribunals capable of reviewing an issued patent are barred from considering. For most AI patents being asserted today, this threshold determination comes in a single forum, and the question arrives earlier and stays alive longer than most litigants plan for.

Congress Barred Eligibility From Nearly Every Tribunal, with One Time-Limited Exception

A petitioner for inter partes review (IPR) cannot challenge subject matter eligibility. 35 U.S.C. §311(b) permits a petitioner to cancel claims "only on a ground that could be raised under section 102 or 103 and only on the basis of prior art consisting of patents or printed publications." There is one notable exception: where a patent owner moves to amend the patent claims, the Board may examine the amended claims on eligibility and written-description and enablement grounds as well. The Federal Circuit decided this directly in American National Manufacturing Inc. v. Sleep Number Corp., 52 F.4th 1371 (Fed. Cir. 2022): "the Board was free to determine whether the proposed claims were unpatentable under §§ 101, 102, 103, and 112." That authority thus arises only when the patent owner volunteers to amend the claims. It does not apply to the issued claims a defendant stands accused of infringing. On prior art, by contrast, inter partes review stays open for the life of the patent, subject to a one-year bar once the challenger has been served with an infringement complaint.

Ex parte reexamination is confined more tightly than the statute alone suggests. A request under 35 U.S.C. §302 must rest on prior art cited under §301, meaning patents and printed publications, not prior use or sale, and §303(a) requires a substantial new question of patentability. The operative limit on eligibility, however, is regulatory. Under 37 C.F.R. §1.552, claims are examined on patents and printed publications, and on §112 only as to subject matter added or deleted during the proceeding. Issues outside that scope "will not be resolved in a reexamination proceeding." A party may raise eligibility, and the examiner will acknowledge it in the next Office action, but the Office will not decide it. Double patenting is the narrow exception, admitted through §303(a)'s reference to "patents and publications," which In re Lonardo, 119 F.3d 960 (Fed. Cir. 1997), read as broader than prior art.

These constraints matter more now than before, as reexamination is where post-grant proceeding volume has increased. Unified Patents reported 594 ex parte reexamination requests in the first half of 2026, up 162.8% from 226 in the first half of 2025, and 74.7% of all post-grant filings. Reexamination requests reached an all-time quarterly high of 336 in the second quarter. Inter partes review moved the other way: 174 petitions in the first half, 57 in the second quarter, an all-time quarterly low, against 201 total PTAB petitions and a 72.5% year-over-year decline. Patently-O noted the recent floor: eleven IPR petitions in the four weeks ending May 2, 2026, the lowest 28-day total since the system opened in September 2012.

Sources attribute the inversion to expanded discretionary denial of IPR petitions, including the "settled expectations" framework and the Director's assumption of personal control over institution decisions, with reexamination held out as the available alternative. The practical consequence is that a validity challenge redirected out of inter partes review now lands in a post-grant proceeding that is even more narrowly confined. The migration added prior-art capacity. It added no authority over eligibility.

Post-grant review (PGR) is the lone exception, as the only Office proceeding that can consider an issued AI patent on all invalidity grounds. Indeed, 35 U.S.C. §321(b) reaches "any ground that could be raised under paragraph (2) or (3) of section 282(b)," which includes eligibility alongside novelty, obviousness, written description, and enablement. But section 321(c) allows a petition only within nine months after grant, and only for patent applications filed under first-inventor-to-file rules (most of them these days). Thus, for a patent that becomes commercially significant years after issuance, the window has closed long before any defendant had a litigation incentive to petition for review. Unified Patents counted 27 post-grant review petitions in the first half of 2026, against 594 reexamination requests.

Two further Office routes can resolve eligibility on issued claims, and neither is available to a challenger. Supplemental examination under 35 U.S.C. §257 is one. It exists mainly to insulate a patent from inequitable-conduct allegations, and its capacity to reach eligibility is a byproduct: 37 C.F.R. §1.625(d)(3) provides that issues beyond patents, printed publications, and amended subject matter "may be considered and resolved, notwithstanding § 1.552(c)." Reissue is examined in the same manner as an original application under 37 C.F.R. §1.176(a), so eligibility is in scope there as well. Both are patent-owner tools. They can be used by the patent owner to address an eligibility problem identified after issuance, but are not available to an accused infringer challenging validity of the asserted patent.

Outside district court and the nine-month post-grant review window, one other adjudicatory forum with limited jurisdiction can still decide eligibility on issued claims: the International Trade Commission (ITC). 19 U.S.C. §1337(c) provides that "[a]ll legal and equitable defenses may be presented in all cases," and a successful eligibility defense defeats a violation. In Certain Polycrystalline Diamond Compacts, Inv. No. 337-TA-1236, the administrative law judge found the asserted claims infringed and not invalid under §§102, 103, and 112, yet ineligible as claiming unpatentable subject matter, and a divided Commission affirmed. The Federal Circuit reversed the eligibility ruling in US Synthetic Corp. v. International Trade Commission, 128 F.4th 1272 (Fed. Cir. 2025), observing that "only the § 101 ruling prevented a section 337 violation." Three constraints keep the ITC route narrow. A Commission patent determination carries no res judicata or collateral estoppel effect in an Article III court, as the Federal Circuit held in Texas Instruments Inc. v. Cypress Semiconductor Corp., 90 F.3d 1558, 1568 (Fed. Cir. 1996), quoting the Trade Reform Act's legislative history. Section 337 reaches only imported articles, and ClearCorrect Operating, LLC v. International Trade Commission, 810 F.3d 1283 (Fed. Cir. 2015), held that "[a]rticles" means "material things" and "does not extend to electronic transmission of digital data," so a model served to United States users from a United States data center likely would fall outside the Commission’s jurisdiction to exclude. No reported investigation involving AI or software has turned on eligibility.

What remains are the federal district courts, whether the patentee files first or an accused infringer files for a declaration of invalidity. The resulting jurisdictional framework produces an asymmetry with real consequences: a patent can clear examination, draw a post-grant challenge, and more often than not emerge unadjudicated on the ground most likely to invalidate the patent once enforced.

AI Patents Draw Patent Office Challenges and Lose in Court

The statistic that captures the mismatch is a finding about where patent challenges are brought. Professor Amy Semet's empirical study of artificial intelligence patent litigation, forthcoming in the Indiana Law Review and discussed here in July, links the USPTO's AI Patent Dataset to district court outcomes across roughly 80,000 litigated patents. As reported in the IP trade press, 23.6% of AI patents asserted in district court are challenged in PTAB proceedings, against 11.9% of asserted non-AI patents. AI patents attract Board challenges at twice the rate, in the tribunal that cannot reach the ground that most often defeats them.

In the district courts, eligibility arrives earlier than any other ground. Of the invalidity determinations in the study, 37.1% involving AI patents came at the pleadings stage, against 9.5% for non-AI patents. The outcomes follow, as that earlier discussion covers in more detail: among cases decided on the merits AI patents were invalidated at 74.2% against 47.9% for non-AI patents, and eligibility accounted for 53.5% of those AI invalidations against 42.0% of the non-AI invalidations. Prof. Semet's methodological limits are noteworthy: the AI classification is probabilistic across confidence thresholds, the findings are associational rather than causal, and each value is conditional on a contested merits disposition, which excludes settlements and default judgments that resolve most patent cases.

One contrary finding is consistent with that screening effect. AI patents are invalidated for obviousness less often than non-AI patents, which Prof. Semet attributes, not to AI claims being more inventive on average, but to cases ending before the fact-intensive obviousness inquiry begins.

MITII Ended Before Discovery

MITII, Inc. v. OpenAI OpCo, LLC, No. 5:26-cv-00191-NW (N.D. Cal. May 8, 2026), is the compressed version of a not-uncommon pattern. MITII, which Judge Noël Wise described as "an intellectual property holding company," asserted four patents against OpenAI's Sora product. The patents describe what MITII characterized as text-to-video technology developed to help children with autism; the lead patent, U.S. 9,397,972, is titled "Animated Delivery of Electronic Messages" and claims text-to-speech conversion with animated character rendering, but not specifically the generative video synthesis techniques performed by frontier text-to-video models.

Venue was an early issue, as the case was filed in the Central District of California in October 2025 and transferred to the Northern District by joint stipulation. A separate Central District order set OpenAI's deadline to file "its motion to dismiss under Rule 12(b)(6) for failure to state a claim because each claim of the asserted patents is directed to patent-ineligible subject matter under 35 U.S.C. § 101" within seven days of docketing in the new district. Venue and schedule were negotiated around a single eligibility motion.

The court treated claim 1 of the '972 patent as representative, as MITII asserted the patent claims collectively in the complaint rather than claim by claim. The court then applied the two-step framework of Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014), which asks first whether a claim is directed to an abstract idea and, if it is, whether the claim recites something more that supplies an inventive concept. At the second step the court found "representative claim 1 is not directed to a specific rule or particular technological solution, but instead is results-oriented and uses a computer as a tool," so that "MITII's alleged inventive steps are abstract ideas themselves." Dismissal was granted without leave to amend, because "[n]o amendment to a complaint can alter what a patent itself states," quoting Sanderling Management Ltd. v. Snap Inc., 65 F.4th 698, 706 (Fed. Cir. 2023). Judgment was entered the same day. There was no claim construction, no contentions, and no prior art in the court record.

GoTV Survived to a Verdict and Fell Anyway

GoTV Streaming, LLC v. Netflix, Inc., Nos. 2024-1669, 2024-1744, 166 F.4th 1053 (Fed. Cir. Feb. 9, 2026) is a similar result — delayed to a much later phase of the case. Netflix raised invalidity for lack of subject matter eligibility on a Rule 12(c) motion for judgment on the pleadings. The Central District of California denied the motion in May 2023 at the first step alone, without reaching the inventive-concept step and without considering the expert declaration GoTV had submitted in opposition to the motion. After claim construction the parties stipulated to entry of judgment against the eligibility challenge, expressly preserving Netflix's rights, and the stipulated judgment was cross-appealed. A three-day trial followed on two claims. The jury found infringement of one of them, claim 4 of U.S. 8,989,715, found no infringement of the asserted '245 claim, and awarded $2.5 million as a life-of-patent lump sum running from the filing of the suit.

The Federal Circuit reversed and directed entry of judgment for Netflix. Judge Taranto's opinion held the representative claim directed to "the abstract idea of a template set of specifications, generic in at least some respects, that can be tailored (in at least one respect) for final production of the specified product (here an image) to fit the user's constraints," and found nothing in the claim to supply an inventive concept at the second step. Because GoTV had made no separate eligibility arguments for the three asserted patents, one representative claim carried all of them. As the IP trade press noted, the court reached the second step in the first instance rather than remanding, and it did so at the parties’ request, neither of which identified factual issues requiring further development.

The disposition was not entirely one-sided. The Federal Circuit also reversed the district court's finding that a claim term common to all claims of the '865 patent was indefinite, and vacated the inducement and damages rulings it did not need to reach. GoTV thus won on one issue of invalidity, but lost the case because lack of eligibility took all the claims down anyway.

An Improvement the Claims Do Not Require Cannot Save Them

A recent example of that principle being applied was in Trustees of Columbia University v. Gen Digital Inc., No. 2024-1243 (Fed. Cir. Mar. 11, 2026), a precedential decision on patents claiming detection of anomalous program execution. The district court, relying primarily on the shared specification, held the claims eligible and struck the eligibility defense before trial. The Federal Circuit held the claims directed to an abstract idea at the first step and vacated the eligibility determination, remanding "for the district court to solely consider, at step two, the question whether the claimed model of function calls feature was conventional."

The Federal Circuit found: "The problem is these supposed improvements are not what the claims are directed to, i.e., the supposed improvements are not required by the language of the asserted claims at all." Two of the reasons generalize to any portfolio. A feature the specification "consistently describes ... as optional" cannot supply eligibility, because on that record "the claims are satisfied even when there is no selective emulation." And a technical feature described in the specification but not required by the claim does not enter the analysis at all. On the underlying subject matter the court invoked Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1319 (Fed. Cir. 2016): "[b]y itself, virus screening is well-known and constitutes an abstract idea."

Two caveats in this case. No claim was held eligible, and none was finally held ineligible either, as the narrow conventionality question remains open on remand. And the opinion never uses the phrase "machine learning," so the decision is relevant to the relationship between claim and specification but is not specifically a machine-learning eligibility holding.

The Inventive-Concept Question Reached a Jury in Ollnova

The eligibility question can also arrive at a stage many litigants do not plan for. In Ollnova Technologies Ltd. v. Ecobee Technologies ULC, Nos. 2025-1045, 2025-1046 (Fed. Cir. June 4, 2026), the Eastern District of Texas found factual disputes precluding summary judgment on eligibility and submitted the inventive-concept step to the jury while instructing that the abstract-idea step was for the court. The Federal Circuit vacated an $11.5 million verdict, holding the instructions and verdict form legally insufficient because they did not identify the abstract idea to which the '495 claims were directed and did not instruct that the abstract idea itself cannot supply the inventive concept. The court affirmed that the claims of two other asserted patents are not directed to an abstract idea.

Taking MITII, GoTV, and Columbia together, the examples provide helpful guidance to plan around. Within the one forum that can consider the ground, eligibility lives at the pleadings, at summary judgment, in a jury instruction, and on appeal. Two of those four cases produced a verdict that was later erased. The forum is singular, but the phase is not.

Only Claimed Features Can Supply Eligibility

The doctrine that connects these cases is narrow and relatively stable. GoTV states it at the outset, quoting ChargePoint, Inc. v. SemaConnect, Inc., 920 F.3d 759, 767 (Fed. Cir. 2019): an ineligibility analysis depends on "the language of the [] [c]laims themselves." The specification and prosecution history inform what the claim is directed to, but "only features that are claimed, not unclaimed details that appear in the specification, can supply something beyond ineligible matter." ChargePoint, 920 F.3d at 769. The same opinion cites AI Visualize, Inc. v. Nuance Communications, Inc., 97 F.4th 1371, 1379 (Fed. Cir. 2024), for the instruction that a court considers the claims in light of the specification while avoiding importing concepts from the specification into the claims. Trading Technologies International, Inc. v. IBG LLC, 921 F.3d 1084, 1095 (Fed. Cir. 2019), states the proposition directly: "Eligibility depends on what is claimed, not all that is disclosed in the specification."

GoTV applied that rule to the labels AI portfolios often rely on. GoTV pointed to the "algorithm" and "architecture" recited in its claim, and the court answered that "those highly general labels, without further identification of details, do not aid GoTV because they can apply to the abstract idea itself." When GoTV argued that specific claim terms required "specific data structures," the court found that label "itself too general," because the terms "refer simply to packages of information of a particular content." The court's summation is instructive: the terms relied on, "notwithstanding any first-blush appearance of technical specificity, have broad meanings that, individually and taken together, cannot support a conclusion that the claim is directed to a concrete computer/network advance."

The Specification Can Supply an Improvement at the Agency but Cannot Add a Missing Claim Limitation in Court

The role of the specification in subject matter eligibility analysis is one area where prosecution practice and litigation practice can diverge in at least some aspects. The patent examination rule appears in MPEP §2106.04(d)(1), reiterated in the USPTO's August 4, 2025 reminders memorandum and in the December 5, 2025 memorandum revising the MPEP in light of Ex parte Desjardins: "[t]he specification need not explicitly set forth the improvement, but it must describe the invention such that the improvement would be apparent to one of ordinary skill in the art." Separately, the claim need not recite the improvement in explicit terms, but must include "the components or steps of the invention that provide the improvement described in the specification." The guidance carries a further requirement that adversaries may latch on to: where the specification asserts an improvement "in a conclusory manner," without the detail necessary to be apparent to a person of ordinary skill, "the examiner should not determine the claim improves technology."

Read together, the examination rule and the litigation rule agree on the decisive requirement: both demand that the claim itself recite the components or steps producing the improvement. The rules diverge on how the claim is read to get there, and that difference shifts the evidentiary burden.

During examination and reexamination, claims are "given their broadest reasonable interpretation consistent with the specification," MPEP §2111, a protocol the Federal Circuit has described as "solely an examination expedient, not a rule of claim construction" and justified because an applicant may still "amend his claims to obtain protection commensurate with his actual contribution to the art." In re Yamamoto, 740 F.2d 1569, 1571 (Fed. Cir. 1984). Under that reading, the specification's role is evidentiary. An examiner may credit an improvement the specification makes apparent to a person of ordinary skill in the art even where the specification never says so expressly, and a declaration can supply that proof.

Issued claims are read under a different standard. District courts, the Commission, and, since a 2018 rule change, the Board in inter partes and post-grant review all apply Phillips v. AWH Corp., 415 F.3d 1303 (Fed. Cir. 2005) (en banc): the ordinary and customary meaning the terms would carry for a person of ordinary skill at the time of the invention, fixed by the claims, the specification, and the prosecution history. MPEP §2111 states the contrast plainly, that "the Office does not interpret claims when examining patent applications in the same manner as the courts," and the Office acknowledged in the 2018 rulemaking that its examination standard "potentially reads on a broader universe of prior art than does the Phillips standard." The choice of standard can decide a case: in PPC Broadband, Inc. v. Corning Optical Communications RF, LLC, 815 F.3d 734, 741 (Fed. Cir. 2016), the Federal Circuit said the standard was "outcome determinative" and that the term at issue would have been construed differently under Phillips.

Thus, the specification's evidentiary function is applied differently in litigation. Once a construction is fixed, the specification has already done its work in setting what the terms mean, and it cannot then be used to add a requirement the construction does not impose. GoTV makes the sequence explicit, taking up eligibility only "[w]ith pertinent claim constructions in hand" and rejecting the patentee's expert testimony because it could not "identify claimed concrete implementations that go beyond the result-focused and functionally described computer processes apparent from the above-recited claim constructions." A patent whose specification describes a technical improvement while its claims recite only a result can therefore issue on the strength of that specification and still fail once its claims are construed.

The Federal Circuit's recent docket confirms which framework governs an issued claim. Beginning at least with US Patent No. 7,679,637 LLC v. Google LLC, No. 2024-1520 (Fed. Cir. Jan. 22, 2026), which affirmed a dismissal for failure to state a claim with leave to amend denied as futile, the court has held software claims ineligible in GoTV and in AGI SureTrack LLC v. Farmers Edge Inc., Nos. 2024-1730, 2024-1830 (Fed. Cir. June 2, 2026), where claims reciting a microprocessor, a GPS receiver, a bus connector, memory, and an application program still failed. It has also vacated findings of eligibility in Columbia and, in part, in Constellation Designs, LLC v. LG Electronics Inc., No. 2024-1822 (Fed. Cir. Apr. 28, 2026, opinion modified Aug. 31, 2026).

Constellation contains a split illustrating both sides of step one of the eligibility analysis. The court divided the appeal into what it called the "optimization claims," which recite a constellation optimized for capacity, and the "constellation claims," which recite specific non-uniform constellations, and held that "this distinction between the result-oriented optimization claims and constellation claims makes all the difference." The optimization claims of the '761 and '700 patents failed because naming the objective was not enough: the representative claim "lacks sufficient details about the alleged 'concrete technique'" for achieving it. The constellation claims of the '509 and '922 patents recited the arrangement itself, including that "the plurality of unique point locations are unequally spaced," that the points "each have a location and a different label," and that "the locations of at least two of the constellation points are the same." Those recited features, the court held, "provide enough specificity and structure to satisfy Alice step one," which meant the claims were not directed to an abstract idea and the court never reached the inventive-concept step at all. Ollnova likewise preserved two patents at the first step. The claims that survived were the claims that recited the mechanism rather than the result.

A Declaration Submitted in Support of Eligibility Can Become Evidence About Claim Scope

The same asymmetry governs the technical declarations becoming more common in AI prosecution, as the USPTO has actively encouraged. The Director's April 30, 2026 best-practices memorandum states that a subject matter eligibility declaration "may demonstrate how one of ordinary skill in the art would interpret a specification that describes a technological improvement," and that "examiners bear the initial burden upon review of all grounds presented under 132 affidavits, which would include SMED submissions, however filed." This can be a real prosecution advantage and, as covered here in July, has moved eligibility practice before the Office from argument toward evidence.

A declaration is also prosecution history, and GoTV shows what a court does with the litigation equivalent. On appeal GoTV relied on its specification and its expert declaration. The Federal Circuit held the testimony "does not, and could not, identify claimed concrete implementations that go beyond the result-focused and functionally described computer processes apparent from the above-recited claim constructions," adding that "conclusory assertions about the speed and efficiency benefits of using ordinary computer and network functionality do not suffice at Alice step two." The court's rule is well-settled and applies to expert declarants by name: "a conclusory assertion about an inventive concept, including by an expert, is insufficient."

The drafting consequences should be carefully considered. A declaration should provide evidence of how a person of ordinary skill in the art would understand what the claims terms require, while avoiding conclusory assertions about what the invention accomplishes. The declaration should also be written on the assumption that a future defendant will scrutinize every sentence for admissions about claim scope, about what was conventional in the prior art, and about the definition of claim terms. The declarant may also become a witness in litigation.

What Companies and Practitioners Can Do Now

  • Audit issued AI claims against the claim-versus-specification question. For each technological improvement the specification emphasizes, identify the claim features required to produce it. Where no claim requires any such feature, the specification cannot carry the eligibility argument, and its description of the improvement can be used to show that the improvement is unclaimed. Consider addressing any gaps in a continuation where one is available, or in a reissue application where necessary, keeping in mind that a broadening reissue must be applied for within two years of grant and remains subject to the recapture rule.
  • Claim sufficient features to accomplish the improvement rather than merely describing it. Reciting generic components like "AI," "algorithm," "neural network," or "architecture" adds nothing when the claim does not specify what those components do differently. The Federal Circuit in GoTV rejected similar labels, and did so as a matter of what the claim terms meant, before reaching whether the labels were inventive.
  • Use eligibility declarations only when arguments and reasonable amendments fail. Ensure the declaration addresses only factors supporting eligibility, avoid statements narrowing the claims beyond desired scope, and confirm the declarant can defend every sentence on cross-examination.
  • Docket the nine-month post-grant review deadline on strategically important patents, including competitors'. Nine months from grant is the only interval in which a USPTO tribunal can test a patent on the full range of invalidity grounds. On a client's own patent that window is a risk period. On a competitor's patent it is the only opportunity to raise eligibility in the Office, and the short window expires without notice.
  • Filing a post-grant review petition requires no stake, but appealing a loss does. Any person other than the patent owner may petition under §321(a), with no requirement of injury, threatened infringement, or competitive position. Article III standing is a separate question. In Incyte Corp. v. Sun Pharmaceutical Industries, Inc., 136 F.4th 1096 (Fed. Cir. 2025), the Federal Circuit dismissed an appeal from a post-grant review because the petitioner's development plans were too speculative to establish a concrete injury. An unthreatened petitioner can therefore lose at the Board with no appellate review available to it.
  • Weigh estoppel and exposure before petitioning against a competitor's patent. Section 325(e)(2) estops a losing petitioner, in district court and before the Commission, from asserting any ground it raised or reasonably could have raised, which in a post-grant review includes eligibility. The petition is also a public filing that names the petitioner and its real parties in interest, tells the patent owner which of its patents the petitioner treats as relevant, and is evidence of the petitioner's knowledge of the patent. A petition therefore suits a party with a documented product or program the claims plainly reach, which both preserves the appeal and makes the trade worthwhile. A petitioner that has already filed a declaratory judgment action challenging validity is barred under §325(a)(1), though §325(a)(3) carves out an invalidity counterclaim.
  • Use available Office proceedings only as part of a complete validity strategy. For an accused infringer, inter partes review and reexamination buy a prior-art review only, while post-grant review is usually already time-barred. Lack of subject matter eligibility is thus often a district-court-only defense, which is why venue and schedule are worth planning around it.
  • Plan for eligibility as a continuing part of the case rather than as a preliminary issue. A patent owner should evaluate eligibility thoroughly before assertion, as MITII illustrates dismissal arriving with prejudice before discovery. A patent owner that succeeds in postponing the question should price the deferral, because GoTV and Ollnova show that a deferred eligibility ruling can arrive only after a verdict, and erase it.

Congress Declined to Add a Second Window in 2011, and No Current Proposal Would Add One

None of these eligibility problems are exclusive to artificial intelligence, and the allocation that creates them was less a considered judgment than an inherited scope and a negotiated one. Ex parte reexamination has been confined to patents and printed publications since Congress created it in 1980, inter partes reexamination took the same scope, and Senator Kyl explained on the floor that the America Invents Act "preserves the agreement reached in the 2009 Judiciary Committee mark up to maintain the current scope of inter partes proceedings: only patents and printed publications may be used to challenge a patent in an inter partes review." 157 Cong. Rec. S1375 (daily ed. Mar. 8, 2011) (statement of Sen. Kyl). The one proceeding that reaches everything else got nine months and a higher institution threshold, the latter "used because some of the issues that can be raised in post-grant review, such as enablement and section 101 invention issues, may require development through discovery." Id. Alice followed three years later and made eligibility the decisive question for exactly the technologies whose share of filings has grown fastest since.

Congress heard the objection this article describes and declined to adopt it. In additional views to the House committee report, Representatives Berman, Watt, and Lofgren wrote that the filing window would "limit the utility of the post-grant review as an efficient, low-cost alternative to litigation," because in several industries "patents are often multitudinous, vague, and highly abstract," which "prevents practitioners from being able to identify and assess relevant patents before they receive some specific warning of liability, which often comes many years after a patent has been issued." They urged a "second window" for post-grant challenges triggered by litigation or a threat of enforcement, as the National Research Council had recommended. H.R. Rep. No. 112-98, pt. 1, at 163-64 (2011) (Additional Views of Reps. Berman, Watt & Lofgren). The proposed second window remains a reasonable fix, yet nothing pending would supply it. The Patent Eligibility Restoration Act of 2025 would rewrite §101 and eliminate the judicial exceptions, but it amends only §§100 and 101, leaving §311(b), the nine-month window, and the forum architecture exactly as they are. Relief from the courts looks no closer: no decision has yet held a machine-learning claim eligible, and the Supreme Court denied certiorari in Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025), in December 2025.

The practical question for a patent issuing today is not what eligibility will mean in five years, but rather that in any given dispute, the claims will still be construed by one tribunal, on one record, on a motion that can be filed at almost any point in the case, while the specification will be unavailable to fill a gap the claims leave open. Prosecution strategy therefore anticipates litigation strategy, and both begin in the same place: claim the technical improvement with sufficient detail to gain allowance and to provide defensibility if the patent is asserted. The claim is the one document every tribunal is permitted to construe.