Senate Debates Whether AI Inventions Can Even Be Patented — and the Stakes Are Enormous

At 10:15 a.m. on July 14, in Room 216 of the Hart Senate Office Building, the Senate Judiciary Committee convened to answer a question that sounds technical but carries enormous economic weight: can artificial intelligence inventions actually be patented under current law, and if not, what happens to the trillions of dollars being invested on the premise that they can?

The hearing — titled “Artificial Intelligence and Intellectual Property, Part I: Patents, Innovation, and Competition” — is the first in a series examining whether the United States patent system is fit for purpose in an era of generative AI. The session arrives against a backdrop of growing alarm in the research and startup communities that existing legal doctrine is systematically stripping AI companies of intellectual property protection at precisely the moment they need it most.

The Alice Problem

To understand why this hearing matters, you need to understand Alice. In 2014, the Supreme Court ruled in Alice Corp. v. CLS Bank International that software implementing an abstract idea could not be patented simply by running it on a computer. The decision was aimed at curbing a wave of trivial software patents that had been used to shakedown legitimate businesses. But its implementation has been far broader and more damaging than intended.

Courts applying the Alice framework — in combination with an earlier ruling, Mayo Collaborative Services v. Prometheus Laboratories — have invalidated patents on everything from diagnostic methods to machine learning algorithms, often on the grounds that they involve an “abstract idea” implemented on generic computing hardware. The test is notoriously vague. Different judges have reached opposite conclusions on nearly identical technology descriptions, and the uncertainty has created a chilling effect on patent prosecution in fields most closely tied to AI.

A study published the day before the hearing by the IP Policy Institute found that AI inventions face significantly higher rates of subject matter eligibility invalidations under Section 101 — the statutory provision where Alice and Mayo operate — compared to patents in other technology categories. The gap is not marginal. Depending on the technical domain, AI-related patents were invalidated at rates two to three times higher than comparable non-AI patents from the same era.

What the Patent Eligibility Restoration Act Would Do

The legislation under discussion, the Patent Eligibility Restoration Act (PERA), would fundamentally rewrite those rules. Its core proposal is the elimination of the judicial exceptions created by Alice and Mayo. Under PERA, the only test for patent eligibility would be whether a claimed invention is in a field of technology and whether it can be practically applied in the real world — a much lower bar than the current framework.

Supporters argue that PERA would restore clarity that US patent law lost in the decade since Alice and create the conditions for American AI companies to protect genuinely novel inventions. They point to the fact that many of the same AI methods that courts have refused to patent in the United States are being actively patented in China and Europe, where patent eligibility doctrine has historically been more permissive toward software and mathematical methods.

Critics counter that PERA overcorrects in dangerous ways. By eliminating the abstract idea exception, the legislation could reopen the door to exactly the kind of low-quality, overbroad software patents that Alice was meant to prevent. Patent trolls — non-practicing entities that acquire patents for the sole purpose of extracting licensing fees — were responsible for more than 60% of all patent litigation in the years before Alice, and many of those cases targeted small tech companies and startups that could not afford to fight back.

A Hearing at the Intersection of Two AI Crises

The timing of the hearing places it at the intersection of two distinct but related pressures on AI intellectual property.

The first is the copyright crisis. AI labs have been sued extensively by authors, artists, news organizations, and record labels over the use of copyrighted material in training datasets. Those cases are working their way through courts and creating uncertainty about what AI outputs themselves can be copyrighted. The patent question is separate but intertwined: even if a training process is legally clear, the resulting model and its applications need IP protection if the company that built them is to have any durable competitive advantage.

The second is the innovation race with China. Chinese AI companies have been filing AI-related patents at roughly three times the rate of US companies for the past several years. While patent count alone is not a reliable measure of innovation quality, the underlying trend — that China’s AI industry is building a formidable IP portfolio while American AI patents face elevated invalidation risk — is the kind of structural asymmetry that tends to concentrate policymaker attention.

Representatives from major research universities, the pharmaceutical industry (whose biotechnology patents face parallel Alice/Mayo problems), the startup community, and civil society all submitted testimony for the hearing. Several witnesses highlighted that the uncertainty created by current doctrine has pushed some AI research toward trade secret protection — keeping algorithms proprietary rather than disclosing them in patent applications — which may slow the diffusion of knowledge that the patent system was designed to encourage.

The Stakes for the AI Investment Cycle

The financial implications are not abstract. Venture capital firms that invest in AI startups typically expect those startups to build a portfolio of patents as part of their defensible moat. If those patents are systematically vulnerable to invalidity challenges under Section 101, a key pillar of the investment thesis is weakened.

Large AI companies are somewhat better insulated: Google, Meta, Microsoft, and OpenAI can protect AI technology through trade secrets, massive compute advantages, and network effects that smaller competitors cannot replicate. But for the mid-tier AI companies — the specialized model developers, the vertical AI SaaS players, the foundation model challengers — the ability to get a patent issued and have it hold up in litigation is often the difference between a fundable business and one that cannot attract institutional capital.

The hearing on July 14 is expected to be the first in a series. The Judiciary Committee has indicated it will examine AI and copyright law in a subsequent session, and there is bipartisan interest in the Senate — unusual in the current political climate — in some form of legislative clarification of AI IP rights, even if PERA itself remains controversial.

The Global Context

The United States is not alone in wrestling with these questions. The European Patent Office has been issuing guidance on AI patentability since 2018, generally taking the position that AI methods applied to a technical process are patentable, while pure mathematical algorithms are not. The UK Intellectual Property Office is conducting a separate review of AI and IP law, expected to produce recommendations later this year. Japan and South Korea have updated their patent examination guidelines to explicitly address AI-generated inventions.

In each jurisdiction, the core tension is the same: the patent system was designed in an era of physical invention, and its application to software and AI methods has been awkward almost from the start. The difference between countries is not whether they face the problem, but how aggressively they choose to solve it.

For the US, where the AI industry is currently the world’s largest and where investor confidence in AI IP is most economically significant, the outcome of the current legislative deliberation will shape the competitive landscape well beyond the current technology cycle. The July 14 hearing was not a vote, not a final decision, and not a resolution. It was, however, the Senate on record acknowledging that the question cannot be deferred much longer.