Intellectual Property Revolution: Modernizing Patent Law for the AI Era

Current landscape: two separate fights

Two separate disputes are live in patent law right now, and conflating them is the easiest way to get this issue wrong. The first is inventorship: does an AI system's contribution to conceiving an invention change who the law recognizes as its inventor? The second is a bipartisan legislative package touching patent eligibility and enforcement that has little to do with AI specifically but happens to be moving through Congress at the same moment AI is straining the patent system's capacity. This issue treats them separately because the evidence for each sits in a different place: court opinions and agency guidance for the first, pending legislation and Patent Office backlog data for the second.

On inventorship, Stephen Thaler spent years trying to list his AI system, DABUS, as sole inventor on patent applications in multiple countries, testing the question directly. The Federal Circuit rejected the U.S. attempt in 2022, holding that the Patent Act's inventorship provisions require a natural person. The Patent Office's first attempt to translate that holding into examination guidance, issued in February 2024, tried to fit AI-assisted invention into the multi-human "joint inventor" test from Pannu v. Iolab Corp. That approach didn't hold up in practice: forcing an AI tool into a test built to compare two or more people produced confusing results for the ordinary case of one human working with one AI system. The Patent Office rescinded and replaced it in November 2025, dropping the Pannu framework and asking a simpler question directly: did a natural person meaningfully direct, select, or interpret the AI's output enough to have conceived the inventive contribution? A related copyright case, Thaler's attempt to copyright AI-generated visual art, reached the same wall: the Supreme Court declined to hear it in March 2026, leaving intact a D.C. Circuit ruling that copyright likewise requires a human author. Patent and copyright law arrived at the same answer through separate doctrines, which is the strongest evidence available that this is a settled question, not an open one.

On eligibility and enforcement, three bipartisan bills are pending: the Patent Eligibility Restoration Act (PERA), which would rewrite Section 101 to eliminate the Alice/Mayo "abstract idea" framework; the PREVAIL Act, which would restrict who can petition the Patent Trial and Appeal Board (PTAB) to challenge a patent's validity; and the RESTORE Patent Rights Act, which would restore a presumption that courts issue injunctions against infringers. All three have the same lead sponsors in the Senate, Thom Tillis (R-NC) and Chris Coons (D-DE), and comparable Republican-Democrat pairs in the House. None has passed as of mid-2026; PREVAIL's 118th Congress predecessor cleared the Senate Judiciary Committee on a narrow 11-10 vote in November 2024 and was reintroduced in the current Congress in May 2025. Supporters, including biotech firms, university tech-transfer offices, and individual-inventor advocates, argue the package restores predictability the courts have eroded. Critics, including the Electronic Frontier Foundation and the startup-focused Engine Advocacy, argue PERA's current text leaves a loophole around its own unmodified-human-gene exclusion (isolating or purifying a gene, a routine step in any genetic test, could count as the "modification" that restores eligibility) alongside near-automatic software eligibility, and that PREVAIL's standing restriction would gut the main tool that keeps non-practicing entities, which already file the large majority of tech-sector patent suits, from extracting settlements from companies too small to litigate.

Proposal mechanics

Proposal 1 answers a durability problem this platform doesn't have to speculate about: it already happened once. The Patent Office's inventorship guidance changed substantively in November 2025, twenty-one months after the prior version issued, with no legislative involvement at all. Guidance a single administration can rewrite is not a stable foundation for billions of dollars in R&D investment decisions. Statute is.

Proposal 2 takes a narrower path than PERA. The Federal Circuit's Recentive Analytics v. Fox Corp. decision, in April 2025, gave Section 101 its first AI-specific test: applying a known machine-learning technique to a new dataset, without a claimed improvement to the underlying model or method, isn't patent-eligible. That standard already distinguishes technical advances from claims that just point an existing tool at a new problem. Writing that standard into statute resolves the uncertainty PERA's supporters are right to flag, chiefly in the software- and AI-heavy technology centers where first-action pendency runs longest, without PERA's broader rewrite of what counts as patentable subject matter at all.

Proposal 3 and Proposal 4 split PREVAIL's enforcement provisions into their separate pieces rather than accepting or rejecting the bill whole. PREVAIL's text already includes a rule barring PTAB from revisiting arguments or prior art it has already considered against a given patent, absent exceptional circumstances. That specific provision, decoupled from PREVAIL's separate restriction on who may petition in the first place, is exactly Proposal 4. Nothing about barring a repeat argument requires also barring the parties, nonprofits, open-source defense groups, and companies that haven't been sued yet but expect to be, who make the first challenge affordable to bring at all.

Proposal 5 addresses capacity rather than doctrine. The Patent Office's Automated Search Pilot Program (ASAP!) uses an internal AI tool to hand examiners and applicants a ranked list of relevant prior art before substantive examination begins. It launched in October 2025 and has already been extended once, through June 2026, with its fee waived and its intake target doubled. A backlog approaching 1.2 million pending applications gets addressed faster by scaling a tool already proving useful than by hoping eligibility litigation sorts itself out.

Trade-offs and opposing views

The strongest objection to Proposals 3 and 4 together is that they don't go as far as PREVAIL's supporters want: a biotech firm or university whose patent survives a first IPR challenge can still, in principle, face a second petition from a different challenger raising different grounds, since Proposal 4 only bars re-litigating the same grounds against the same patent, not all further challenge. That's a gap relative to PREVAIL's more sweeping deference language, accepted deliberately: PREVAIL's broader deference provisions are difficult to separate from the standing restriction this issue rejects without also insulating patents from challenges that raise new prior art, which is the entire point of a validity review process. The strongest objection to Proposal 2 is the mirror image: some AI and diagnostic claims that PERA's supporters believe deserve protection will still fail the technical-improvement standard this issue proposes instead. That is answered directly in Who Bears the Cost, below, rather than waved off here.